xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa.c (revision 4f0f5e5be9d3811b437d9156675d584e2a2f204a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This file contains all the routines used when modifying on-disk SPA state.
29  * This includes opening, importing, destroying, exporting a pool, and syncing a
30  * pool.
31  */
32 
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/zio.h>
37 #include <sys/zio_checksum.h>
38 #include <sys/zio_compress.h>
39 #include <sys/dmu.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/zap.h>
42 #include <sys/zil.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/uberblock_impl.h>
46 #include <sys/txg.h>
47 #include <sys/avl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/arc.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/sunddi.h>
61 #include <sys/spa_boot.h>
62 #include <sys/zfs_ioctl.h>
63 
64 #ifdef	_KERNEL
65 #include <sys/zone.h>
66 #endif	/* _KERNEL */
67 
68 #include "zfs_prop.h"
69 #include "zfs_comutil.h"
70 
71 enum zti_modes {
72 	zti_mode_fixed,			/* value is # of threads (min 1) */
73 	zti_mode_online_percent,	/* value is % of online CPUs */
74 	zti_mode_tune,			/* fill from zio_taskq_tune_* */
75 	zti_nmodes
76 };
77 
78 #define	ZTI_THREAD_FIX(n)	{ zti_mode_fixed, (n) }
79 #define	ZTI_THREAD_PCT(n)	{ zti_mode_online_percent, (n) }
80 #define	ZTI_THREAD_TUNE		{ zti_mode_tune, 0 }
81 
82 #define	ZTI_THREAD_ONE		ZTI_THREAD_FIX(1)
83 
84 typedef struct zio_taskq_info {
85 	const char *zti_name;
86 	struct {
87 		enum zti_modes zti_mode;
88 		uint_t zti_value;
89 	} zti_nthreads[ZIO_TASKQ_TYPES];
90 } zio_taskq_info_t;
91 
92 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
93 				"issue",		"intr"
94 };
95 
96 const zio_taskq_info_t zio_taskqs[ZIO_TYPES] = {
97 	/*			ISSUE			INTR		*/
98 	{ "spa_zio_null",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
99 	{ "spa_zio_read",	{ ZTI_THREAD_FIX(8),	ZTI_THREAD_TUNE } },
100 	{ "spa_zio_write",	{ ZTI_THREAD_TUNE,	ZTI_THREAD_FIX(8) } },
101 	{ "spa_zio_free",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
102 	{ "spa_zio_claim",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
103 	{ "spa_zio_ioctl",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
104 };
105 
106 enum zti_modes zio_taskq_tune_mode = zti_mode_online_percent;
107 uint_t zio_taskq_tune_value = 80;	/* #threads = 80% of # online CPUs */
108 
109 static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
110 static boolean_t spa_has_active_shared_spare(spa_t *spa);
111 
112 /*
113  * ==========================================================================
114  * SPA properties routines
115  * ==========================================================================
116  */
117 
118 /*
119  * Add a (source=src, propname=propval) list to an nvlist.
120  */
121 static void
122 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
123     uint64_t intval, zprop_source_t src)
124 {
125 	const char *propname = zpool_prop_to_name(prop);
126 	nvlist_t *propval;
127 
128 	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
129 	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
130 
131 	if (strval != NULL)
132 		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
133 	else
134 		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
135 
136 	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
137 	nvlist_free(propval);
138 }
139 
140 /*
141  * Get property values from the spa configuration.
142  */
143 static void
144 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
145 {
146 	uint64_t size;
147 	uint64_t used;
148 	uint64_t cap, version;
149 	zprop_source_t src = ZPROP_SRC_NONE;
150 	spa_config_dirent_t *dp;
151 
152 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
153 
154 	if (spa->spa_root_vdev != NULL) {
155 		size = spa_get_space(spa);
156 		used = spa_get_alloc(spa);
157 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
158 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
159 		spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
160 		spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL,
161 		    size - used, src);
162 
163 		cap = (size == 0) ? 0 : (used * 100 / size);
164 		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
165 
166 		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
167 		    spa->spa_root_vdev->vdev_state, src);
168 
169 		version = spa_version(spa);
170 		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
171 			src = ZPROP_SRC_DEFAULT;
172 		else
173 			src = ZPROP_SRC_LOCAL;
174 		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
175 	}
176 
177 	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
178 
179 	if (spa->spa_root != NULL)
180 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
181 		    0, ZPROP_SRC_LOCAL);
182 
183 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
184 		if (dp->scd_path == NULL) {
185 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
186 			    "none", 0, ZPROP_SRC_LOCAL);
187 		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
188 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
189 			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
190 		}
191 	}
192 }
193 
194 /*
195  * Get zpool property values.
196  */
197 int
198 spa_prop_get(spa_t *spa, nvlist_t **nvp)
199 {
200 	zap_cursor_t zc;
201 	zap_attribute_t za;
202 	objset_t *mos = spa->spa_meta_objset;
203 	int err;
204 
205 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
206 
207 	mutex_enter(&spa->spa_props_lock);
208 
209 	/*
210 	 * Get properties from the spa config.
211 	 */
212 	spa_prop_get_config(spa, nvp);
213 
214 	/* If no pool property object, no more prop to get. */
215 	if (spa->spa_pool_props_object == 0) {
216 		mutex_exit(&spa->spa_props_lock);
217 		return (0);
218 	}
219 
220 	/*
221 	 * Get properties from the MOS pool property object.
222 	 */
223 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
224 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
225 	    zap_cursor_advance(&zc)) {
226 		uint64_t intval = 0;
227 		char *strval = NULL;
228 		zprop_source_t src = ZPROP_SRC_DEFAULT;
229 		zpool_prop_t prop;
230 
231 		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
232 			continue;
233 
234 		switch (za.za_integer_length) {
235 		case 8:
236 			/* integer property */
237 			if (za.za_first_integer !=
238 			    zpool_prop_default_numeric(prop))
239 				src = ZPROP_SRC_LOCAL;
240 
241 			if (prop == ZPOOL_PROP_BOOTFS) {
242 				dsl_pool_t *dp;
243 				dsl_dataset_t *ds = NULL;
244 
245 				dp = spa_get_dsl(spa);
246 				rw_enter(&dp->dp_config_rwlock, RW_READER);
247 				if (err = dsl_dataset_hold_obj(dp,
248 				    za.za_first_integer, FTAG, &ds)) {
249 					rw_exit(&dp->dp_config_rwlock);
250 					break;
251 				}
252 
253 				strval = kmem_alloc(
254 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
255 				    KM_SLEEP);
256 				dsl_dataset_name(ds, strval);
257 				dsl_dataset_rele(ds, FTAG);
258 				rw_exit(&dp->dp_config_rwlock);
259 			} else {
260 				strval = NULL;
261 				intval = za.za_first_integer;
262 			}
263 
264 			spa_prop_add_list(*nvp, prop, strval, intval, src);
265 
266 			if (strval != NULL)
267 				kmem_free(strval,
268 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
269 
270 			break;
271 
272 		case 1:
273 			/* string property */
274 			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
275 			err = zap_lookup(mos, spa->spa_pool_props_object,
276 			    za.za_name, 1, za.za_num_integers, strval);
277 			if (err) {
278 				kmem_free(strval, za.za_num_integers);
279 				break;
280 			}
281 			spa_prop_add_list(*nvp, prop, strval, 0, src);
282 			kmem_free(strval, za.za_num_integers);
283 			break;
284 
285 		default:
286 			break;
287 		}
288 	}
289 	zap_cursor_fini(&zc);
290 	mutex_exit(&spa->spa_props_lock);
291 out:
292 	if (err && err != ENOENT) {
293 		nvlist_free(*nvp);
294 		*nvp = NULL;
295 		return (err);
296 	}
297 
298 	return (0);
299 }
300 
301 /*
302  * Validate the given pool properties nvlist and modify the list
303  * for the property values to be set.
304  */
305 static int
306 spa_prop_validate(spa_t *spa, nvlist_t *props)
307 {
308 	nvpair_t *elem;
309 	int error = 0, reset_bootfs = 0;
310 	uint64_t objnum;
311 
312 	elem = NULL;
313 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
314 		zpool_prop_t prop;
315 		char *propname, *strval;
316 		uint64_t intval;
317 		objset_t *os;
318 		char *slash;
319 
320 		propname = nvpair_name(elem);
321 
322 		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
323 			return (EINVAL);
324 
325 		switch (prop) {
326 		case ZPOOL_PROP_VERSION:
327 			error = nvpair_value_uint64(elem, &intval);
328 			if (!error &&
329 			    (intval < spa_version(spa) || intval > SPA_VERSION))
330 				error = EINVAL;
331 			break;
332 
333 		case ZPOOL_PROP_DELEGATION:
334 		case ZPOOL_PROP_AUTOREPLACE:
335 		case ZPOOL_PROP_LISTSNAPS:
336 		case ZPOOL_PROP_AUTOEXPAND:
337 			error = nvpair_value_uint64(elem, &intval);
338 			if (!error && intval > 1)
339 				error = EINVAL;
340 			break;
341 
342 		case ZPOOL_PROP_BOOTFS:
343 			/*
344 			 * If the pool version is less than SPA_VERSION_BOOTFS,
345 			 * or the pool is still being created (version == 0),
346 			 * the bootfs property cannot be set.
347 			 */
348 			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
349 				error = ENOTSUP;
350 				break;
351 			}
352 
353 			/*
354 			 * Make sure the vdev config is bootable
355 			 */
356 			if (!vdev_is_bootable(spa->spa_root_vdev)) {
357 				error = ENOTSUP;
358 				break;
359 			}
360 
361 			reset_bootfs = 1;
362 
363 			error = nvpair_value_string(elem, &strval);
364 
365 			if (!error) {
366 				uint64_t compress;
367 
368 				if (strval == NULL || strval[0] == '\0') {
369 					objnum = zpool_prop_default_numeric(
370 					    ZPOOL_PROP_BOOTFS);
371 					break;
372 				}
373 
374 				if (error = dmu_objset_open(strval, DMU_OST_ZFS,
375 				    DS_MODE_USER | DS_MODE_READONLY, &os))
376 					break;
377 
378 				/* We don't support gzip bootable datasets */
379 				if ((error = dsl_prop_get_integer(strval,
380 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
381 				    &compress, NULL)) == 0 &&
382 				    !BOOTFS_COMPRESS_VALID(compress)) {
383 					error = ENOTSUP;
384 				} else {
385 					objnum = dmu_objset_id(os);
386 				}
387 				dmu_objset_close(os);
388 			}
389 			break;
390 
391 		case ZPOOL_PROP_FAILUREMODE:
392 			error = nvpair_value_uint64(elem, &intval);
393 			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
394 			    intval > ZIO_FAILURE_MODE_PANIC))
395 				error = EINVAL;
396 
397 			/*
398 			 * This is a special case which only occurs when
399 			 * the pool has completely failed. This allows
400 			 * the user to change the in-core failmode property
401 			 * without syncing it out to disk (I/Os might
402 			 * currently be blocked). We do this by returning
403 			 * EIO to the caller (spa_prop_set) to trick it
404 			 * into thinking we encountered a property validation
405 			 * error.
406 			 */
407 			if (!error && spa_suspended(spa)) {
408 				spa->spa_failmode = intval;
409 				error = EIO;
410 			}
411 			break;
412 
413 		case ZPOOL_PROP_CACHEFILE:
414 			if ((error = nvpair_value_string(elem, &strval)) != 0)
415 				break;
416 
417 			if (strval[0] == '\0')
418 				break;
419 
420 			if (strcmp(strval, "none") == 0)
421 				break;
422 
423 			if (strval[0] != '/') {
424 				error = EINVAL;
425 				break;
426 			}
427 
428 			slash = strrchr(strval, '/');
429 			ASSERT(slash != NULL);
430 
431 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
432 			    strcmp(slash, "/..") == 0)
433 				error = EINVAL;
434 			break;
435 		}
436 
437 		if (error)
438 			break;
439 	}
440 
441 	if (!error && reset_bootfs) {
442 		error = nvlist_remove(props,
443 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
444 
445 		if (!error) {
446 			error = nvlist_add_uint64(props,
447 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
448 		}
449 	}
450 
451 	return (error);
452 }
453 
454 void
455 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
456 {
457 	char *cachefile;
458 	spa_config_dirent_t *dp;
459 
460 	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
461 	    &cachefile) != 0)
462 		return;
463 
464 	dp = kmem_alloc(sizeof (spa_config_dirent_t),
465 	    KM_SLEEP);
466 
467 	if (cachefile[0] == '\0')
468 		dp->scd_path = spa_strdup(spa_config_path);
469 	else if (strcmp(cachefile, "none") == 0)
470 		dp->scd_path = NULL;
471 	else
472 		dp->scd_path = spa_strdup(cachefile);
473 
474 	list_insert_head(&spa->spa_config_list, dp);
475 	if (need_sync)
476 		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
477 }
478 
479 int
480 spa_prop_set(spa_t *spa, nvlist_t *nvp)
481 {
482 	int error;
483 	nvpair_t *elem;
484 	boolean_t need_sync = B_FALSE;
485 	zpool_prop_t prop;
486 
487 	if ((error = spa_prop_validate(spa, nvp)) != 0)
488 		return (error);
489 
490 	elem = NULL;
491 	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
492 		if ((prop = zpool_name_to_prop(
493 		    nvpair_name(elem))) == ZPROP_INVAL)
494 			return (EINVAL);
495 
496 		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
497 			continue;
498 
499 		need_sync = B_TRUE;
500 		break;
501 	}
502 
503 	if (need_sync)
504 		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
505 		    spa, nvp, 3));
506 	else
507 		return (0);
508 }
509 
510 /*
511  * If the bootfs property value is dsobj, clear it.
512  */
513 void
514 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
515 {
516 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
517 		VERIFY(zap_remove(spa->spa_meta_objset,
518 		    spa->spa_pool_props_object,
519 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
520 		spa->spa_bootfs = 0;
521 	}
522 }
523 
524 /*
525  * ==========================================================================
526  * SPA state manipulation (open/create/destroy/import/export)
527  * ==========================================================================
528  */
529 
530 static int
531 spa_error_entry_compare(const void *a, const void *b)
532 {
533 	spa_error_entry_t *sa = (spa_error_entry_t *)a;
534 	spa_error_entry_t *sb = (spa_error_entry_t *)b;
535 	int ret;
536 
537 	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
538 	    sizeof (zbookmark_t));
539 
540 	if (ret < 0)
541 		return (-1);
542 	else if (ret > 0)
543 		return (1);
544 	else
545 		return (0);
546 }
547 
548 /*
549  * Utility function which retrieves copies of the current logs and
550  * re-initializes them in the process.
551  */
552 void
553 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
554 {
555 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
556 
557 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
558 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
559 
560 	avl_create(&spa->spa_errlist_scrub,
561 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
562 	    offsetof(spa_error_entry_t, se_avl));
563 	avl_create(&spa->spa_errlist_last,
564 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
565 	    offsetof(spa_error_entry_t, se_avl));
566 }
567 
568 /*
569  * Activate an uninitialized pool.
570  */
571 static void
572 spa_activate(spa_t *spa, int mode)
573 {
574 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
575 
576 	spa->spa_state = POOL_STATE_ACTIVE;
577 	spa->spa_mode = mode;
578 
579 	spa->spa_normal_class = metaslab_class_create(zfs_metaslab_ops);
580 	spa->spa_log_class = metaslab_class_create(zfs_metaslab_ops);
581 
582 	for (int t = 0; t < ZIO_TYPES; t++) {
583 		const zio_taskq_info_t *ztip = &zio_taskqs[t];
584 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
585 			enum zti_modes mode = ztip->zti_nthreads[q].zti_mode;
586 			uint_t value = ztip->zti_nthreads[q].zti_value;
587 			char name[32];
588 
589 			(void) snprintf(name, sizeof (name),
590 			    "%s_%s", ztip->zti_name, zio_taskq_types[q]);
591 
592 			if (mode == zti_mode_tune) {
593 				mode = zio_taskq_tune_mode;
594 				value = zio_taskq_tune_value;
595 				if (mode == zti_mode_tune)
596 					mode = zti_mode_online_percent;
597 			}
598 
599 			switch (mode) {
600 			case zti_mode_fixed:
601 				ASSERT3U(value, >=, 1);
602 				value = MAX(value, 1);
603 
604 				spa->spa_zio_taskq[t][q] = taskq_create(name,
605 				    value, maxclsyspri, 50, INT_MAX,
606 				    TASKQ_PREPOPULATE);
607 				break;
608 
609 			case zti_mode_online_percent:
610 				spa->spa_zio_taskq[t][q] = taskq_create(name,
611 				    value, maxclsyspri, 50, INT_MAX,
612 				    TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);
613 				break;
614 
615 			case zti_mode_tune:
616 			default:
617 				panic("unrecognized mode for "
618 				    "zio_taskqs[%u]->zti_nthreads[%u] (%u:%u) "
619 				    "in spa_activate()",
620 				    t, q, mode, value);
621 				break;
622 			}
623 		}
624 	}
625 
626 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
627 	    offsetof(vdev_t, vdev_config_dirty_node));
628 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
629 	    offsetof(vdev_t, vdev_state_dirty_node));
630 
631 	txg_list_create(&spa->spa_vdev_txg_list,
632 	    offsetof(struct vdev, vdev_txg_node));
633 
634 	avl_create(&spa->spa_errlist_scrub,
635 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
636 	    offsetof(spa_error_entry_t, se_avl));
637 	avl_create(&spa->spa_errlist_last,
638 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
639 	    offsetof(spa_error_entry_t, se_avl));
640 }
641 
642 /*
643  * Opposite of spa_activate().
644  */
645 static void
646 spa_deactivate(spa_t *spa)
647 {
648 	ASSERT(spa->spa_sync_on == B_FALSE);
649 	ASSERT(spa->spa_dsl_pool == NULL);
650 	ASSERT(spa->spa_root_vdev == NULL);
651 	ASSERT(spa->spa_async_zio_root == NULL);
652 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
653 
654 	txg_list_destroy(&spa->spa_vdev_txg_list);
655 
656 	list_destroy(&spa->spa_config_dirty_list);
657 	list_destroy(&spa->spa_state_dirty_list);
658 
659 	for (int t = 0; t < ZIO_TYPES; t++) {
660 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
661 			taskq_destroy(spa->spa_zio_taskq[t][q]);
662 			spa->spa_zio_taskq[t][q] = NULL;
663 		}
664 	}
665 
666 	metaslab_class_destroy(spa->spa_normal_class);
667 	spa->spa_normal_class = NULL;
668 
669 	metaslab_class_destroy(spa->spa_log_class);
670 	spa->spa_log_class = NULL;
671 
672 	/*
673 	 * If this was part of an import or the open otherwise failed, we may
674 	 * still have errors left in the queues.  Empty them just in case.
675 	 */
676 	spa_errlog_drain(spa);
677 
678 	avl_destroy(&spa->spa_errlist_scrub);
679 	avl_destroy(&spa->spa_errlist_last);
680 
681 	spa->spa_state = POOL_STATE_UNINITIALIZED;
682 }
683 
684 /*
685  * Verify a pool configuration, and construct the vdev tree appropriately.  This
686  * will create all the necessary vdevs in the appropriate layout, with each vdev
687  * in the CLOSED state.  This will prep the pool before open/creation/import.
688  * All vdev validation is done by the vdev_alloc() routine.
689  */
690 static int
691 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
692     uint_t id, int atype)
693 {
694 	nvlist_t **child;
695 	uint_t children;
696 	int error;
697 
698 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
699 		return (error);
700 
701 	if ((*vdp)->vdev_ops->vdev_op_leaf)
702 		return (0);
703 
704 	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
705 	    &child, &children);
706 
707 	if (error == ENOENT)
708 		return (0);
709 
710 	if (error) {
711 		vdev_free(*vdp);
712 		*vdp = NULL;
713 		return (EINVAL);
714 	}
715 
716 	for (int c = 0; c < children; c++) {
717 		vdev_t *vd;
718 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
719 		    atype)) != 0) {
720 			vdev_free(*vdp);
721 			*vdp = NULL;
722 			return (error);
723 		}
724 	}
725 
726 	ASSERT(*vdp != NULL);
727 
728 	return (0);
729 }
730 
731 /*
732  * Opposite of spa_load().
733  */
734 static void
735 spa_unload(spa_t *spa)
736 {
737 	int i;
738 
739 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
740 
741 	/*
742 	 * Stop async tasks.
743 	 */
744 	spa_async_suspend(spa);
745 
746 	/*
747 	 * Stop syncing.
748 	 */
749 	if (spa->spa_sync_on) {
750 		txg_sync_stop(spa->spa_dsl_pool);
751 		spa->spa_sync_on = B_FALSE;
752 	}
753 
754 	/*
755 	 * Wait for any outstanding async I/O to complete.
756 	 */
757 	if (spa->spa_async_zio_root != NULL) {
758 		(void) zio_wait(spa->spa_async_zio_root);
759 		spa->spa_async_zio_root = NULL;
760 	}
761 
762 	/*
763 	 * Close the dsl pool.
764 	 */
765 	if (spa->spa_dsl_pool) {
766 		dsl_pool_close(spa->spa_dsl_pool);
767 		spa->spa_dsl_pool = NULL;
768 	}
769 
770 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
771 
772 	/*
773 	 * Drop and purge level 2 cache
774 	 */
775 	spa_l2cache_drop(spa);
776 
777 	/*
778 	 * Close all vdevs.
779 	 */
780 	if (spa->spa_root_vdev)
781 		vdev_free(spa->spa_root_vdev);
782 	ASSERT(spa->spa_root_vdev == NULL);
783 
784 	for (i = 0; i < spa->spa_spares.sav_count; i++)
785 		vdev_free(spa->spa_spares.sav_vdevs[i]);
786 	if (spa->spa_spares.sav_vdevs) {
787 		kmem_free(spa->spa_spares.sav_vdevs,
788 		    spa->spa_spares.sav_count * sizeof (void *));
789 		spa->spa_spares.sav_vdevs = NULL;
790 	}
791 	if (spa->spa_spares.sav_config) {
792 		nvlist_free(spa->spa_spares.sav_config);
793 		spa->spa_spares.sav_config = NULL;
794 	}
795 	spa->spa_spares.sav_count = 0;
796 
797 	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
798 		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
799 	if (spa->spa_l2cache.sav_vdevs) {
800 		kmem_free(spa->spa_l2cache.sav_vdevs,
801 		    spa->spa_l2cache.sav_count * sizeof (void *));
802 		spa->spa_l2cache.sav_vdevs = NULL;
803 	}
804 	if (spa->spa_l2cache.sav_config) {
805 		nvlist_free(spa->spa_l2cache.sav_config);
806 		spa->spa_l2cache.sav_config = NULL;
807 	}
808 	spa->spa_l2cache.sav_count = 0;
809 
810 	spa->spa_async_suspended = 0;
811 
812 	spa_config_exit(spa, SCL_ALL, FTAG);
813 }
814 
815 /*
816  * Load (or re-load) the current list of vdevs describing the active spares for
817  * this pool.  When this is called, we have some form of basic information in
818  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
819  * then re-generate a more complete list including status information.
820  */
821 static void
822 spa_load_spares(spa_t *spa)
823 {
824 	nvlist_t **spares;
825 	uint_t nspares;
826 	int i;
827 	vdev_t *vd, *tvd;
828 
829 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
830 
831 	/*
832 	 * First, close and free any existing spare vdevs.
833 	 */
834 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
835 		vd = spa->spa_spares.sav_vdevs[i];
836 
837 		/* Undo the call to spa_activate() below */
838 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
839 		    B_FALSE)) != NULL && tvd->vdev_isspare)
840 			spa_spare_remove(tvd);
841 		vdev_close(vd);
842 		vdev_free(vd);
843 	}
844 
845 	if (spa->spa_spares.sav_vdevs)
846 		kmem_free(spa->spa_spares.sav_vdevs,
847 		    spa->spa_spares.sav_count * sizeof (void *));
848 
849 	if (spa->spa_spares.sav_config == NULL)
850 		nspares = 0;
851 	else
852 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
853 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
854 
855 	spa->spa_spares.sav_count = (int)nspares;
856 	spa->spa_spares.sav_vdevs = NULL;
857 
858 	if (nspares == 0)
859 		return;
860 
861 	/*
862 	 * Construct the array of vdevs, opening them to get status in the
863 	 * process.   For each spare, there is potentially two different vdev_t
864 	 * structures associated with it: one in the list of spares (used only
865 	 * for basic validation purposes) and one in the active vdev
866 	 * configuration (if it's spared in).  During this phase we open and
867 	 * validate each vdev on the spare list.  If the vdev also exists in the
868 	 * active configuration, then we also mark this vdev as an active spare.
869 	 */
870 	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
871 	    KM_SLEEP);
872 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
873 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
874 		    VDEV_ALLOC_SPARE) == 0);
875 		ASSERT(vd != NULL);
876 
877 		spa->spa_spares.sav_vdevs[i] = vd;
878 
879 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
880 		    B_FALSE)) != NULL) {
881 			if (!tvd->vdev_isspare)
882 				spa_spare_add(tvd);
883 
884 			/*
885 			 * We only mark the spare active if we were successfully
886 			 * able to load the vdev.  Otherwise, importing a pool
887 			 * with a bad active spare would result in strange
888 			 * behavior, because multiple pool would think the spare
889 			 * is actively in use.
890 			 *
891 			 * There is a vulnerability here to an equally bizarre
892 			 * circumstance, where a dead active spare is later
893 			 * brought back to life (onlined or otherwise).  Given
894 			 * the rarity of this scenario, and the extra complexity
895 			 * it adds, we ignore the possibility.
896 			 */
897 			if (!vdev_is_dead(tvd))
898 				spa_spare_activate(tvd);
899 		}
900 
901 		vd->vdev_top = vd;
902 		vd->vdev_aux = &spa->spa_spares;
903 
904 		if (vdev_open(vd) != 0)
905 			continue;
906 
907 		if (vdev_validate_aux(vd) == 0)
908 			spa_spare_add(vd);
909 	}
910 
911 	/*
912 	 * Recompute the stashed list of spares, with status information
913 	 * this time.
914 	 */
915 	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
916 	    DATA_TYPE_NVLIST_ARRAY) == 0);
917 
918 	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
919 	    KM_SLEEP);
920 	for (i = 0; i < spa->spa_spares.sav_count; i++)
921 		spares[i] = vdev_config_generate(spa,
922 		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
923 	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
924 	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
925 	for (i = 0; i < spa->spa_spares.sav_count; i++)
926 		nvlist_free(spares[i]);
927 	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
928 }
929 
930 /*
931  * Load (or re-load) the current list of vdevs describing the active l2cache for
932  * this pool.  When this is called, we have some form of basic information in
933  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
934  * then re-generate a more complete list including status information.
935  * Devices which are already active have their details maintained, and are
936  * not re-opened.
937  */
938 static void
939 spa_load_l2cache(spa_t *spa)
940 {
941 	nvlist_t **l2cache;
942 	uint_t nl2cache;
943 	int i, j, oldnvdevs;
944 	uint64_t guid;
945 	vdev_t *vd, **oldvdevs, **newvdevs;
946 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
947 
948 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
949 
950 	if (sav->sav_config != NULL) {
951 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
952 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
953 		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
954 	} else {
955 		nl2cache = 0;
956 	}
957 
958 	oldvdevs = sav->sav_vdevs;
959 	oldnvdevs = sav->sav_count;
960 	sav->sav_vdevs = NULL;
961 	sav->sav_count = 0;
962 
963 	/*
964 	 * Process new nvlist of vdevs.
965 	 */
966 	for (i = 0; i < nl2cache; i++) {
967 		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
968 		    &guid) == 0);
969 
970 		newvdevs[i] = NULL;
971 		for (j = 0; j < oldnvdevs; j++) {
972 			vd = oldvdevs[j];
973 			if (vd != NULL && guid == vd->vdev_guid) {
974 				/*
975 				 * Retain previous vdev for add/remove ops.
976 				 */
977 				newvdevs[i] = vd;
978 				oldvdevs[j] = NULL;
979 				break;
980 			}
981 		}
982 
983 		if (newvdevs[i] == NULL) {
984 			/*
985 			 * Create new vdev
986 			 */
987 			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
988 			    VDEV_ALLOC_L2CACHE) == 0);
989 			ASSERT(vd != NULL);
990 			newvdevs[i] = vd;
991 
992 			/*
993 			 * Commit this vdev as an l2cache device,
994 			 * even if it fails to open.
995 			 */
996 			spa_l2cache_add(vd);
997 
998 			vd->vdev_top = vd;
999 			vd->vdev_aux = sav;
1000 
1001 			spa_l2cache_activate(vd);
1002 
1003 			if (vdev_open(vd) != 0)
1004 				continue;
1005 
1006 			(void) vdev_validate_aux(vd);
1007 
1008 			if (!vdev_is_dead(vd))
1009 				l2arc_add_vdev(spa, vd);
1010 		}
1011 	}
1012 
1013 	/*
1014 	 * Purge vdevs that were dropped
1015 	 */
1016 	for (i = 0; i < oldnvdevs; i++) {
1017 		uint64_t pool;
1018 
1019 		vd = oldvdevs[i];
1020 		if (vd != NULL) {
1021 			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1022 			    pool != 0ULL && l2arc_vdev_present(vd))
1023 				l2arc_remove_vdev(vd);
1024 			(void) vdev_close(vd);
1025 			spa_l2cache_remove(vd);
1026 		}
1027 	}
1028 
1029 	if (oldvdevs)
1030 		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1031 
1032 	if (sav->sav_config == NULL)
1033 		goto out;
1034 
1035 	sav->sav_vdevs = newvdevs;
1036 	sav->sav_count = (int)nl2cache;
1037 
1038 	/*
1039 	 * Recompute the stashed list of l2cache devices, with status
1040 	 * information this time.
1041 	 */
1042 	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1043 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1044 
1045 	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1046 	for (i = 0; i < sav->sav_count; i++)
1047 		l2cache[i] = vdev_config_generate(spa,
1048 		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
1049 	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1050 	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1051 out:
1052 	for (i = 0; i < sav->sav_count; i++)
1053 		nvlist_free(l2cache[i]);
1054 	if (sav->sav_count)
1055 		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1056 }
1057 
1058 static int
1059 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1060 {
1061 	dmu_buf_t *db;
1062 	char *packed = NULL;
1063 	size_t nvsize = 0;
1064 	int error;
1065 	*value = NULL;
1066 
1067 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1068 	nvsize = *(uint64_t *)db->db_data;
1069 	dmu_buf_rele(db, FTAG);
1070 
1071 	packed = kmem_alloc(nvsize, KM_SLEEP);
1072 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1073 	    DMU_READ_PREFETCH);
1074 	if (error == 0)
1075 		error = nvlist_unpack(packed, nvsize, value, 0);
1076 	kmem_free(packed, nvsize);
1077 
1078 	return (error);
1079 }
1080 
1081 /*
1082  * Checks to see if the given vdev could not be opened, in which case we post a
1083  * sysevent to notify the autoreplace code that the device has been removed.
1084  */
1085 static void
1086 spa_check_removed(vdev_t *vd)
1087 {
1088 	for (int c = 0; c < vd->vdev_children; c++)
1089 		spa_check_removed(vd->vdev_child[c]);
1090 
1091 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1092 		zfs_post_autoreplace(vd->vdev_spa, vd);
1093 		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1094 	}
1095 }
1096 
1097 /*
1098  * Load the slog device state from the config object since it's possible
1099  * that the label does not contain the most up-to-date information.
1100  */
1101 void
1102 spa_load_log_state(spa_t *spa)
1103 {
1104 	nvlist_t *nv, *nvroot, **child;
1105 	uint64_t is_log;
1106 	uint_t children;
1107 	vdev_t *rvd = spa->spa_root_vdev;
1108 
1109 	VERIFY(load_nvlist(spa, spa->spa_config_object, &nv) == 0);
1110 	VERIFY(nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1111 	VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1112 	    &child, &children) == 0);
1113 
1114 	for (int c = 0; c < children; c++) {
1115 		vdev_t *tvd = rvd->vdev_child[c];
1116 
1117 		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1118 		    &is_log) == 0 && is_log)
1119 			vdev_load_log_state(tvd, child[c]);
1120 	}
1121 	nvlist_free(nv);
1122 }
1123 
1124 /*
1125  * Check for missing log devices
1126  */
1127 int
1128 spa_check_logs(spa_t *spa)
1129 {
1130 	switch (spa->spa_log_state) {
1131 	case SPA_LOG_MISSING:
1132 		/* need to recheck in case slog has been restored */
1133 	case SPA_LOG_UNKNOWN:
1134 		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1135 		    DS_FIND_CHILDREN)) {
1136 			spa->spa_log_state = SPA_LOG_MISSING;
1137 			return (1);
1138 		}
1139 		break;
1140 	}
1141 	return (0);
1142 }
1143 
1144 /*
1145  * Load an existing storage pool, using the pool's builtin spa_config as a
1146  * source of configuration information.
1147  */
1148 static int
1149 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1150 {
1151 	int error = 0;
1152 	nvlist_t *nvroot = NULL;
1153 	vdev_t *rvd;
1154 	uberblock_t *ub = &spa->spa_uberblock;
1155 	uint64_t config_cache_txg = spa->spa_config_txg;
1156 	uint64_t pool_guid;
1157 	uint64_t version;
1158 	uint64_t autoreplace = 0;
1159 	int orig_mode = spa->spa_mode;
1160 	char *ereport = FM_EREPORT_ZFS_POOL;
1161 
1162 	/*
1163 	 * If this is an untrusted config, access the pool in read-only mode.
1164 	 * This prevents things like resilvering recently removed devices.
1165 	 */
1166 	if (!mosconfig)
1167 		spa->spa_mode = FREAD;
1168 
1169 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1170 
1171 	spa->spa_load_state = state;
1172 
1173 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1174 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1175 		error = EINVAL;
1176 		goto out;
1177 	}
1178 
1179 	/*
1180 	 * Versioning wasn't explicitly added to the label until later, so if
1181 	 * it's not present treat it as the initial version.
1182 	 */
1183 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1184 		version = SPA_VERSION_INITIAL;
1185 
1186 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1187 	    &spa->spa_config_txg);
1188 
1189 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1190 	    spa_guid_exists(pool_guid, 0)) {
1191 		error = EEXIST;
1192 		goto out;
1193 	}
1194 
1195 	spa->spa_load_guid = pool_guid;
1196 
1197 	/*
1198 	 * Create "The Godfather" zio to hold all async IOs
1199 	 */
1200 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1201 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1202 
1203 	/*
1204 	 * Parse the configuration into a vdev tree.  We explicitly set the
1205 	 * value that will be returned by spa_version() since parsing the
1206 	 * configuration requires knowing the version number.
1207 	 */
1208 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1209 	spa->spa_ubsync.ub_version = version;
1210 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1211 	spa_config_exit(spa, SCL_ALL, FTAG);
1212 
1213 	if (error != 0)
1214 		goto out;
1215 
1216 	ASSERT(spa->spa_root_vdev == rvd);
1217 	ASSERT(spa_guid(spa) == pool_guid);
1218 
1219 	/*
1220 	 * Try to open all vdevs, loading each label in the process.
1221 	 */
1222 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1223 	error = vdev_open(rvd);
1224 	spa_config_exit(spa, SCL_ALL, FTAG);
1225 	if (error != 0)
1226 		goto out;
1227 
1228 	/*
1229 	 * We need to validate the vdev labels against the configuration that
1230 	 * we have in hand, which is dependent on the setting of mosconfig. If
1231 	 * mosconfig is true then we're validating the vdev labels based on
1232 	 * that config. Otherwise, we're validating against the cached config
1233 	 * (zpool.cache) that was read when we loaded the zfs module, and then
1234 	 * later we will recursively call spa_load() and validate against
1235 	 * the vdev config.
1236 	 */
1237 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1238 	error = vdev_validate(rvd);
1239 	spa_config_exit(spa, SCL_ALL, FTAG);
1240 	if (error != 0)
1241 		goto out;
1242 
1243 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1244 		error = ENXIO;
1245 		goto out;
1246 	}
1247 
1248 	/*
1249 	 * Find the best uberblock.
1250 	 */
1251 	vdev_uberblock_load(NULL, rvd, ub);
1252 
1253 	/*
1254 	 * If we weren't able to find a single valid uberblock, return failure.
1255 	 */
1256 	if (ub->ub_txg == 0) {
1257 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1258 		    VDEV_AUX_CORRUPT_DATA);
1259 		error = ENXIO;
1260 		goto out;
1261 	}
1262 
1263 	/*
1264 	 * If the pool is newer than the code, we can't open it.
1265 	 */
1266 	if (ub->ub_version > SPA_VERSION) {
1267 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1268 		    VDEV_AUX_VERSION_NEWER);
1269 		error = ENOTSUP;
1270 		goto out;
1271 	}
1272 
1273 	/*
1274 	 * If the vdev guid sum doesn't match the uberblock, we have an
1275 	 * incomplete configuration.
1276 	 */
1277 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1278 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1279 		    VDEV_AUX_BAD_GUID_SUM);
1280 		error = ENXIO;
1281 		goto out;
1282 	}
1283 
1284 	/*
1285 	 * Initialize internal SPA structures.
1286 	 */
1287 	spa->spa_state = POOL_STATE_ACTIVE;
1288 	spa->spa_ubsync = spa->spa_uberblock;
1289 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1290 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1291 	if (error) {
1292 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1293 		    VDEV_AUX_CORRUPT_DATA);
1294 		goto out;
1295 	}
1296 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1297 
1298 	if (zap_lookup(spa->spa_meta_objset,
1299 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1300 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1301 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1302 		    VDEV_AUX_CORRUPT_DATA);
1303 		error = EIO;
1304 		goto out;
1305 	}
1306 
1307 	if (!mosconfig) {
1308 		nvlist_t *newconfig;
1309 		uint64_t hostid;
1310 
1311 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1312 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1313 			    VDEV_AUX_CORRUPT_DATA);
1314 			error = EIO;
1315 			goto out;
1316 		}
1317 
1318 		if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1319 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1320 			char *hostname;
1321 			unsigned long myhostid = 0;
1322 
1323 			VERIFY(nvlist_lookup_string(newconfig,
1324 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1325 
1326 #ifdef	_KERNEL
1327 			myhostid = zone_get_hostid(NULL);
1328 #else	/* _KERNEL */
1329 			/*
1330 			 * We're emulating the system's hostid in userland, so
1331 			 * we can't use zone_get_hostid().
1332 			 */
1333 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1334 #endif	/* _KERNEL */
1335 			if (hostid != 0 && myhostid != 0 &&
1336 			    hostid != myhostid) {
1337 				cmn_err(CE_WARN, "pool '%s' could not be "
1338 				    "loaded as it was last accessed by "
1339 				    "another system (host: %s hostid: 0x%lx). "
1340 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1341 				    spa_name(spa), hostname,
1342 				    (unsigned long)hostid);
1343 				error = EBADF;
1344 				goto out;
1345 			}
1346 		}
1347 
1348 		spa_config_set(spa, newconfig);
1349 		spa_unload(spa);
1350 		spa_deactivate(spa);
1351 		spa_activate(spa, orig_mode);
1352 
1353 		return (spa_load(spa, newconfig, state, B_TRUE));
1354 	}
1355 
1356 	if (zap_lookup(spa->spa_meta_objset,
1357 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1358 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1359 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1360 		    VDEV_AUX_CORRUPT_DATA);
1361 		error = EIO;
1362 		goto out;
1363 	}
1364 
1365 	/*
1366 	 * Load the bit that tells us to use the new accounting function
1367 	 * (raid-z deflation).  If we have an older pool, this will not
1368 	 * be present.
1369 	 */
1370 	error = zap_lookup(spa->spa_meta_objset,
1371 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1372 	    sizeof (uint64_t), 1, &spa->spa_deflate);
1373 	if (error != 0 && error != ENOENT) {
1374 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1375 		    VDEV_AUX_CORRUPT_DATA);
1376 		error = EIO;
1377 		goto out;
1378 	}
1379 
1380 	/*
1381 	 * Load the persistent error log.  If we have an older pool, this will
1382 	 * not be present.
1383 	 */
1384 	error = zap_lookup(spa->spa_meta_objset,
1385 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1386 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1387 	if (error != 0 && error != ENOENT) {
1388 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1389 		    VDEV_AUX_CORRUPT_DATA);
1390 		error = EIO;
1391 		goto out;
1392 	}
1393 
1394 	error = zap_lookup(spa->spa_meta_objset,
1395 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1396 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1397 	if (error != 0 && error != ENOENT) {
1398 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1399 		    VDEV_AUX_CORRUPT_DATA);
1400 		error = EIO;
1401 		goto out;
1402 	}
1403 
1404 	/*
1405 	 * Load the history object.  If we have an older pool, this
1406 	 * will not be present.
1407 	 */
1408 	error = zap_lookup(spa->spa_meta_objset,
1409 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1410 	    sizeof (uint64_t), 1, &spa->spa_history);
1411 	if (error != 0 && error != ENOENT) {
1412 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1413 		    VDEV_AUX_CORRUPT_DATA);
1414 		error = EIO;
1415 		goto out;
1416 	}
1417 
1418 	/*
1419 	 * Load any hot spares for this pool.
1420 	 */
1421 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1422 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1423 	if (error != 0 && error != ENOENT) {
1424 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1425 		    VDEV_AUX_CORRUPT_DATA);
1426 		error = EIO;
1427 		goto out;
1428 	}
1429 	if (error == 0) {
1430 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1431 		if (load_nvlist(spa, spa->spa_spares.sav_object,
1432 		    &spa->spa_spares.sav_config) != 0) {
1433 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1434 			    VDEV_AUX_CORRUPT_DATA);
1435 			error = EIO;
1436 			goto out;
1437 		}
1438 
1439 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1440 		spa_load_spares(spa);
1441 		spa_config_exit(spa, SCL_ALL, FTAG);
1442 	}
1443 
1444 	/*
1445 	 * Load any level 2 ARC devices for this pool.
1446 	 */
1447 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1448 	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1449 	    &spa->spa_l2cache.sav_object);
1450 	if (error != 0 && error != ENOENT) {
1451 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1452 		    VDEV_AUX_CORRUPT_DATA);
1453 		error = EIO;
1454 		goto out;
1455 	}
1456 	if (error == 0) {
1457 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1458 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1459 		    &spa->spa_l2cache.sav_config) != 0) {
1460 			vdev_set_state(rvd, B_TRUE,
1461 			    VDEV_STATE_CANT_OPEN,
1462 			    VDEV_AUX_CORRUPT_DATA);
1463 			error = EIO;
1464 			goto out;
1465 		}
1466 
1467 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1468 		spa_load_l2cache(spa);
1469 		spa_config_exit(spa, SCL_ALL, FTAG);
1470 	}
1471 
1472 	spa_load_log_state(spa);
1473 
1474 	if (spa_check_logs(spa)) {
1475 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1476 		    VDEV_AUX_BAD_LOG);
1477 		error = ENXIO;
1478 		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1479 		goto out;
1480 	}
1481 
1482 
1483 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1484 
1485 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1486 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1487 
1488 	if (error && error != ENOENT) {
1489 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1490 		    VDEV_AUX_CORRUPT_DATA);
1491 		error = EIO;
1492 		goto out;
1493 	}
1494 
1495 	if (error == 0) {
1496 		(void) zap_lookup(spa->spa_meta_objset,
1497 		    spa->spa_pool_props_object,
1498 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1499 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1500 		(void) zap_lookup(spa->spa_meta_objset,
1501 		    spa->spa_pool_props_object,
1502 		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1503 		    sizeof (uint64_t), 1, &autoreplace);
1504 		(void) zap_lookup(spa->spa_meta_objset,
1505 		    spa->spa_pool_props_object,
1506 		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1507 		    sizeof (uint64_t), 1, &spa->spa_delegation);
1508 		(void) zap_lookup(spa->spa_meta_objset,
1509 		    spa->spa_pool_props_object,
1510 		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1511 		    sizeof (uint64_t), 1, &spa->spa_failmode);
1512 		(void) zap_lookup(spa->spa_meta_objset,
1513 		    spa->spa_pool_props_object,
1514 		    zpool_prop_to_name(ZPOOL_PROP_AUTOEXPAND),
1515 		    sizeof (uint64_t), 1, &spa->spa_autoexpand);
1516 	}
1517 
1518 	/*
1519 	 * If the 'autoreplace' property is set, then post a resource notifying
1520 	 * the ZFS DE that it should not issue any faults for unopenable
1521 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1522 	 * unopenable vdevs so that the normal autoreplace handler can take
1523 	 * over.
1524 	 */
1525 	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1526 		spa_check_removed(spa->spa_root_vdev);
1527 
1528 	/*
1529 	 * Load the vdev state for all toplevel vdevs.
1530 	 */
1531 	vdev_load(rvd);
1532 
1533 	/*
1534 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1535 	 */
1536 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1537 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1538 	spa_config_exit(spa, SCL_ALL, FTAG);
1539 
1540 	/*
1541 	 * Check the state of the root vdev.  If it can't be opened, it
1542 	 * indicates one or more toplevel vdevs are faulted.
1543 	 */
1544 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1545 		error = ENXIO;
1546 		goto out;
1547 	}
1548 
1549 	if (spa_writeable(spa)) {
1550 		dmu_tx_t *tx;
1551 		int need_update = B_FALSE;
1552 
1553 		ASSERT(state != SPA_LOAD_TRYIMPORT);
1554 
1555 		/*
1556 		 * Claim log blocks that haven't been committed yet.
1557 		 * This must all happen in a single txg.
1558 		 */
1559 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1560 		    spa_first_txg(spa));
1561 		(void) dmu_objset_find(spa_name(spa),
1562 		    zil_claim, tx, DS_FIND_CHILDREN);
1563 		dmu_tx_commit(tx);
1564 
1565 		spa->spa_log_state = SPA_LOG_GOOD;
1566 		spa->spa_sync_on = B_TRUE;
1567 		txg_sync_start(spa->spa_dsl_pool);
1568 
1569 		/*
1570 		 * Wait for all claims to sync.
1571 		 */
1572 		txg_wait_synced(spa->spa_dsl_pool, 0);
1573 
1574 		/*
1575 		 * If the config cache is stale, or we have uninitialized
1576 		 * metaslabs (see spa_vdev_add()), then update the config.
1577 		 */
1578 		if (config_cache_txg != spa->spa_config_txg ||
1579 		    state == SPA_LOAD_IMPORT)
1580 			need_update = B_TRUE;
1581 
1582 		for (int c = 0; c < rvd->vdev_children; c++)
1583 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1584 				need_update = B_TRUE;
1585 
1586 		/*
1587 		 * Update the config cache asychronously in case we're the
1588 		 * root pool, in which case the config cache isn't writable yet.
1589 		 */
1590 		if (need_update)
1591 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1592 
1593 		/*
1594 		 * Check all DTLs to see if anything needs resilvering.
1595 		 */
1596 		if (vdev_resilver_needed(rvd, NULL, NULL))
1597 			spa_async_request(spa, SPA_ASYNC_RESILVER);
1598 	}
1599 
1600 	error = 0;
1601 out:
1602 	spa->spa_minref = refcount_count(&spa->spa_refcount);
1603 	if (error && error != EBADF)
1604 		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1605 	spa->spa_load_state = SPA_LOAD_NONE;
1606 	spa->spa_ena = 0;
1607 
1608 	return (error);
1609 }
1610 
1611 /*
1612  * Pool Open/Import
1613  *
1614  * The import case is identical to an open except that the configuration is sent
1615  * down from userland, instead of grabbed from the configuration cache.  For the
1616  * case of an open, the pool configuration will exist in the
1617  * POOL_STATE_UNINITIALIZED state.
1618  *
1619  * The stats information (gen/count/ustats) is used to gather vdev statistics at
1620  * the same time open the pool, without having to keep around the spa_t in some
1621  * ambiguous state.
1622  */
1623 static int
1624 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1625 {
1626 	spa_t *spa;
1627 	int error;
1628 	int locked = B_FALSE;
1629 
1630 	*spapp = NULL;
1631 
1632 	/*
1633 	 * As disgusting as this is, we need to support recursive calls to this
1634 	 * function because dsl_dir_open() is called during spa_load(), and ends
1635 	 * up calling spa_open() again.  The real fix is to figure out how to
1636 	 * avoid dsl_dir_open() calling this in the first place.
1637 	 */
1638 	if (mutex_owner(&spa_namespace_lock) != curthread) {
1639 		mutex_enter(&spa_namespace_lock);
1640 		locked = B_TRUE;
1641 	}
1642 
1643 	if ((spa = spa_lookup(pool)) == NULL) {
1644 		if (locked)
1645 			mutex_exit(&spa_namespace_lock);
1646 		return (ENOENT);
1647 	}
1648 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1649 
1650 		spa_activate(spa, spa_mode_global);
1651 
1652 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1653 
1654 		if (error == EBADF) {
1655 			/*
1656 			 * If vdev_validate() returns failure (indicated by
1657 			 * EBADF), it indicates that one of the vdevs indicates
1658 			 * that the pool has been exported or destroyed.  If
1659 			 * this is the case, the config cache is out of sync and
1660 			 * we should remove the pool from the namespace.
1661 			 */
1662 			spa_unload(spa);
1663 			spa_deactivate(spa);
1664 			spa_config_sync(spa, B_TRUE, B_TRUE);
1665 			spa_remove(spa);
1666 			if (locked)
1667 				mutex_exit(&spa_namespace_lock);
1668 			return (ENOENT);
1669 		}
1670 
1671 		if (error) {
1672 			/*
1673 			 * We can't open the pool, but we still have useful
1674 			 * information: the state of each vdev after the
1675 			 * attempted vdev_open().  Return this to the user.
1676 			 */
1677 			if (config != NULL && spa->spa_root_vdev != NULL)
1678 				*config = spa_config_generate(spa, NULL, -1ULL,
1679 				    B_TRUE);
1680 			spa_unload(spa);
1681 			spa_deactivate(spa);
1682 			spa->spa_last_open_failed = B_TRUE;
1683 			if (locked)
1684 				mutex_exit(&spa_namespace_lock);
1685 			*spapp = NULL;
1686 			return (error);
1687 		} else {
1688 			spa->spa_last_open_failed = B_FALSE;
1689 		}
1690 	}
1691 
1692 	spa_open_ref(spa, tag);
1693 
1694 	if (locked)
1695 		mutex_exit(&spa_namespace_lock);
1696 
1697 	*spapp = spa;
1698 
1699 	if (config != NULL)
1700 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1701 
1702 	return (0);
1703 }
1704 
1705 int
1706 spa_open(const char *name, spa_t **spapp, void *tag)
1707 {
1708 	return (spa_open_common(name, spapp, tag, NULL));
1709 }
1710 
1711 /*
1712  * Lookup the given spa_t, incrementing the inject count in the process,
1713  * preventing it from being exported or destroyed.
1714  */
1715 spa_t *
1716 spa_inject_addref(char *name)
1717 {
1718 	spa_t *spa;
1719 
1720 	mutex_enter(&spa_namespace_lock);
1721 	if ((spa = spa_lookup(name)) == NULL) {
1722 		mutex_exit(&spa_namespace_lock);
1723 		return (NULL);
1724 	}
1725 	spa->spa_inject_ref++;
1726 	mutex_exit(&spa_namespace_lock);
1727 
1728 	return (spa);
1729 }
1730 
1731 void
1732 spa_inject_delref(spa_t *spa)
1733 {
1734 	mutex_enter(&spa_namespace_lock);
1735 	spa->spa_inject_ref--;
1736 	mutex_exit(&spa_namespace_lock);
1737 }
1738 
1739 /*
1740  * Add spares device information to the nvlist.
1741  */
1742 static void
1743 spa_add_spares(spa_t *spa, nvlist_t *config)
1744 {
1745 	nvlist_t **spares;
1746 	uint_t i, nspares;
1747 	nvlist_t *nvroot;
1748 	uint64_t guid;
1749 	vdev_stat_t *vs;
1750 	uint_t vsc;
1751 	uint64_t pool;
1752 
1753 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1754 
1755 	if (spa->spa_spares.sav_count == 0)
1756 		return;
1757 
1758 	VERIFY(nvlist_lookup_nvlist(config,
1759 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1760 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1761 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1762 	if (nspares != 0) {
1763 		VERIFY(nvlist_add_nvlist_array(nvroot,
1764 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1765 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1766 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1767 
1768 		/*
1769 		 * Go through and find any spares which have since been
1770 		 * repurposed as an active spare.  If this is the case, update
1771 		 * their status appropriately.
1772 		 */
1773 		for (i = 0; i < nspares; i++) {
1774 			VERIFY(nvlist_lookup_uint64(spares[i],
1775 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1776 			if (spa_spare_exists(guid, &pool, NULL) &&
1777 			    pool != 0ULL) {
1778 				VERIFY(nvlist_lookup_uint64_array(
1779 				    spares[i], ZPOOL_CONFIG_STATS,
1780 				    (uint64_t **)&vs, &vsc) == 0);
1781 				vs->vs_state = VDEV_STATE_CANT_OPEN;
1782 				vs->vs_aux = VDEV_AUX_SPARED;
1783 			}
1784 		}
1785 	}
1786 }
1787 
1788 /*
1789  * Add l2cache device information to the nvlist, including vdev stats.
1790  */
1791 static void
1792 spa_add_l2cache(spa_t *spa, nvlist_t *config)
1793 {
1794 	nvlist_t **l2cache;
1795 	uint_t i, j, nl2cache;
1796 	nvlist_t *nvroot;
1797 	uint64_t guid;
1798 	vdev_t *vd;
1799 	vdev_stat_t *vs;
1800 	uint_t vsc;
1801 
1802 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1803 
1804 	if (spa->spa_l2cache.sav_count == 0)
1805 		return;
1806 
1807 	VERIFY(nvlist_lookup_nvlist(config,
1808 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1809 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1810 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1811 	if (nl2cache != 0) {
1812 		VERIFY(nvlist_add_nvlist_array(nvroot,
1813 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1814 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1815 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1816 
1817 		/*
1818 		 * Update level 2 cache device stats.
1819 		 */
1820 
1821 		for (i = 0; i < nl2cache; i++) {
1822 			VERIFY(nvlist_lookup_uint64(l2cache[i],
1823 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1824 
1825 			vd = NULL;
1826 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1827 				if (guid ==
1828 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1829 					vd = spa->spa_l2cache.sav_vdevs[j];
1830 					break;
1831 				}
1832 			}
1833 			ASSERT(vd != NULL);
1834 
1835 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1836 			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1837 			vdev_get_stats(vd, vs);
1838 		}
1839 	}
1840 }
1841 
1842 int
1843 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1844 {
1845 	int error;
1846 	spa_t *spa;
1847 
1848 	*config = NULL;
1849 	error = spa_open_common(name, &spa, FTAG, config);
1850 
1851 	if (spa != NULL) {
1852 		/*
1853 		 * This still leaves a window of inconsistency where the spares
1854 		 * or l2cache devices could change and the config would be
1855 		 * self-inconsistent.
1856 		 */
1857 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1858 
1859 		if (*config != NULL) {
1860 			VERIFY(nvlist_add_uint64(*config,
1861 			    ZPOOL_CONFIG_ERRCOUNT,
1862 			    spa_get_errlog_size(spa)) == 0);
1863 
1864 			if (spa_suspended(spa))
1865 				VERIFY(nvlist_add_uint64(*config,
1866 				    ZPOOL_CONFIG_SUSPENDED,
1867 				    spa->spa_failmode) == 0);
1868 
1869 			spa_add_spares(spa, *config);
1870 			spa_add_l2cache(spa, *config);
1871 		}
1872 	}
1873 
1874 	/*
1875 	 * We want to get the alternate root even for faulted pools, so we cheat
1876 	 * and call spa_lookup() directly.
1877 	 */
1878 	if (altroot) {
1879 		if (spa == NULL) {
1880 			mutex_enter(&spa_namespace_lock);
1881 			spa = spa_lookup(name);
1882 			if (spa)
1883 				spa_altroot(spa, altroot, buflen);
1884 			else
1885 				altroot[0] = '\0';
1886 			spa = NULL;
1887 			mutex_exit(&spa_namespace_lock);
1888 		} else {
1889 			spa_altroot(spa, altroot, buflen);
1890 		}
1891 	}
1892 
1893 	if (spa != NULL) {
1894 		spa_config_exit(spa, SCL_CONFIG, FTAG);
1895 		spa_close(spa, FTAG);
1896 	}
1897 
1898 	return (error);
1899 }
1900 
1901 /*
1902  * Validate that the auxiliary device array is well formed.  We must have an
1903  * array of nvlists, each which describes a valid leaf vdev.  If this is an
1904  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1905  * specified, as long as they are well-formed.
1906  */
1907 static int
1908 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1909     spa_aux_vdev_t *sav, const char *config, uint64_t version,
1910     vdev_labeltype_t label)
1911 {
1912 	nvlist_t **dev;
1913 	uint_t i, ndev;
1914 	vdev_t *vd;
1915 	int error;
1916 
1917 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1918 
1919 	/*
1920 	 * It's acceptable to have no devs specified.
1921 	 */
1922 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1923 		return (0);
1924 
1925 	if (ndev == 0)
1926 		return (EINVAL);
1927 
1928 	/*
1929 	 * Make sure the pool is formatted with a version that supports this
1930 	 * device type.
1931 	 */
1932 	if (spa_version(spa) < version)
1933 		return (ENOTSUP);
1934 
1935 	/*
1936 	 * Set the pending device list so we correctly handle device in-use
1937 	 * checking.
1938 	 */
1939 	sav->sav_pending = dev;
1940 	sav->sav_npending = ndev;
1941 
1942 	for (i = 0; i < ndev; i++) {
1943 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1944 		    mode)) != 0)
1945 			goto out;
1946 
1947 		if (!vd->vdev_ops->vdev_op_leaf) {
1948 			vdev_free(vd);
1949 			error = EINVAL;
1950 			goto out;
1951 		}
1952 
1953 		/*
1954 		 * The L2ARC currently only supports disk devices in
1955 		 * kernel context.  For user-level testing, we allow it.
1956 		 */
1957 #ifdef _KERNEL
1958 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1959 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1960 			error = ENOTBLK;
1961 			goto out;
1962 		}
1963 #endif
1964 		vd->vdev_top = vd;
1965 
1966 		if ((error = vdev_open(vd)) == 0 &&
1967 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1968 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1969 			    vd->vdev_guid) == 0);
1970 		}
1971 
1972 		vdev_free(vd);
1973 
1974 		if (error &&
1975 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1976 			goto out;
1977 		else
1978 			error = 0;
1979 	}
1980 
1981 out:
1982 	sav->sav_pending = NULL;
1983 	sav->sav_npending = 0;
1984 	return (error);
1985 }
1986 
1987 static int
1988 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1989 {
1990 	int error;
1991 
1992 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1993 
1994 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1995 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1996 	    VDEV_LABEL_SPARE)) != 0) {
1997 		return (error);
1998 	}
1999 
2000 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2001 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2002 	    VDEV_LABEL_L2CACHE));
2003 }
2004 
2005 static void
2006 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2007     const char *config)
2008 {
2009 	int i;
2010 
2011 	if (sav->sav_config != NULL) {
2012 		nvlist_t **olddevs;
2013 		uint_t oldndevs;
2014 		nvlist_t **newdevs;
2015 
2016 		/*
2017 		 * Generate new dev list by concatentating with the
2018 		 * current dev list.
2019 		 */
2020 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2021 		    &olddevs, &oldndevs) == 0);
2022 
2023 		newdevs = kmem_alloc(sizeof (void *) *
2024 		    (ndevs + oldndevs), KM_SLEEP);
2025 		for (i = 0; i < oldndevs; i++)
2026 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2027 			    KM_SLEEP) == 0);
2028 		for (i = 0; i < ndevs; i++)
2029 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2030 			    KM_SLEEP) == 0);
2031 
2032 		VERIFY(nvlist_remove(sav->sav_config, config,
2033 		    DATA_TYPE_NVLIST_ARRAY) == 0);
2034 
2035 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2036 		    config, newdevs, ndevs + oldndevs) == 0);
2037 		for (i = 0; i < oldndevs + ndevs; i++)
2038 			nvlist_free(newdevs[i]);
2039 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2040 	} else {
2041 		/*
2042 		 * Generate a new dev list.
2043 		 */
2044 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2045 		    KM_SLEEP) == 0);
2046 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2047 		    devs, ndevs) == 0);
2048 	}
2049 }
2050 
2051 /*
2052  * Stop and drop level 2 ARC devices
2053  */
2054 void
2055 spa_l2cache_drop(spa_t *spa)
2056 {
2057 	vdev_t *vd;
2058 	int i;
2059 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2060 
2061 	for (i = 0; i < sav->sav_count; i++) {
2062 		uint64_t pool;
2063 
2064 		vd = sav->sav_vdevs[i];
2065 		ASSERT(vd != NULL);
2066 
2067 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2068 		    pool != 0ULL && l2arc_vdev_present(vd))
2069 			l2arc_remove_vdev(vd);
2070 		if (vd->vdev_isl2cache)
2071 			spa_l2cache_remove(vd);
2072 		vdev_clear_stats(vd);
2073 		(void) vdev_close(vd);
2074 	}
2075 }
2076 
2077 /*
2078  * Pool Creation
2079  */
2080 int
2081 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2082     const char *history_str, nvlist_t *zplprops)
2083 {
2084 	spa_t *spa;
2085 	char *altroot = NULL;
2086 	vdev_t *rvd;
2087 	dsl_pool_t *dp;
2088 	dmu_tx_t *tx;
2089 	int error = 0;
2090 	uint64_t txg = TXG_INITIAL;
2091 	nvlist_t **spares, **l2cache;
2092 	uint_t nspares, nl2cache;
2093 	uint64_t version;
2094 
2095 	/*
2096 	 * If this pool already exists, return failure.
2097 	 */
2098 	mutex_enter(&spa_namespace_lock);
2099 	if (spa_lookup(pool) != NULL) {
2100 		mutex_exit(&spa_namespace_lock);
2101 		return (EEXIST);
2102 	}
2103 
2104 	/*
2105 	 * Allocate a new spa_t structure.
2106 	 */
2107 	(void) nvlist_lookup_string(props,
2108 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2109 	spa = spa_add(pool, altroot);
2110 	spa_activate(spa, spa_mode_global);
2111 
2112 	spa->spa_uberblock.ub_txg = txg - 1;
2113 
2114 	if (props && (error = spa_prop_validate(spa, props))) {
2115 		spa_deactivate(spa);
2116 		spa_remove(spa);
2117 		mutex_exit(&spa_namespace_lock);
2118 		return (error);
2119 	}
2120 
2121 	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2122 	    &version) != 0)
2123 		version = SPA_VERSION;
2124 	ASSERT(version <= SPA_VERSION);
2125 	spa->spa_uberblock.ub_version = version;
2126 	spa->spa_ubsync = spa->spa_uberblock;
2127 
2128 	/*
2129 	 * Create "The Godfather" zio to hold all async IOs
2130 	 */
2131 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2132 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2133 
2134 	/*
2135 	 * Create the root vdev.
2136 	 */
2137 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2138 
2139 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2140 
2141 	ASSERT(error != 0 || rvd != NULL);
2142 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2143 
2144 	if (error == 0 && !zfs_allocatable_devs(nvroot))
2145 		error = EINVAL;
2146 
2147 	if (error == 0 &&
2148 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2149 	    (error = spa_validate_aux(spa, nvroot, txg,
2150 	    VDEV_ALLOC_ADD)) == 0) {
2151 		for (int c = 0; c < rvd->vdev_children; c++) {
2152 			vdev_metaslab_set_size(rvd->vdev_child[c]);
2153 			vdev_expand(rvd->vdev_child[c], txg);
2154 		}
2155 	}
2156 
2157 	spa_config_exit(spa, SCL_ALL, FTAG);
2158 
2159 	if (error != 0) {
2160 		spa_unload(spa);
2161 		spa_deactivate(spa);
2162 		spa_remove(spa);
2163 		mutex_exit(&spa_namespace_lock);
2164 		return (error);
2165 	}
2166 
2167 	/*
2168 	 * Get the list of spares, if specified.
2169 	 */
2170 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2171 	    &spares, &nspares) == 0) {
2172 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2173 		    KM_SLEEP) == 0);
2174 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2175 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2176 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2177 		spa_load_spares(spa);
2178 		spa_config_exit(spa, SCL_ALL, FTAG);
2179 		spa->spa_spares.sav_sync = B_TRUE;
2180 	}
2181 
2182 	/*
2183 	 * Get the list of level 2 cache devices, if specified.
2184 	 */
2185 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2186 	    &l2cache, &nl2cache) == 0) {
2187 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2188 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2189 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2190 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2191 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2192 		spa_load_l2cache(spa);
2193 		spa_config_exit(spa, SCL_ALL, FTAG);
2194 		spa->spa_l2cache.sav_sync = B_TRUE;
2195 	}
2196 
2197 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2198 	spa->spa_meta_objset = dp->dp_meta_objset;
2199 
2200 	tx = dmu_tx_create_assigned(dp, txg);
2201 
2202 	/*
2203 	 * Create the pool config object.
2204 	 */
2205 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2206 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2207 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2208 
2209 	if (zap_add(spa->spa_meta_objset,
2210 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2211 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2212 		cmn_err(CE_PANIC, "failed to add pool config");
2213 	}
2214 
2215 	/* Newly created pools with the right version are always deflated. */
2216 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2217 		spa->spa_deflate = TRUE;
2218 		if (zap_add(spa->spa_meta_objset,
2219 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2220 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2221 			cmn_err(CE_PANIC, "failed to add deflate");
2222 		}
2223 	}
2224 
2225 	/*
2226 	 * Create the deferred-free bplist object.  Turn off compression
2227 	 * because sync-to-convergence takes longer if the blocksize
2228 	 * keeps changing.
2229 	 */
2230 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2231 	    1 << 14, tx);
2232 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2233 	    ZIO_COMPRESS_OFF, tx);
2234 
2235 	if (zap_add(spa->spa_meta_objset,
2236 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2237 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2238 		cmn_err(CE_PANIC, "failed to add bplist");
2239 	}
2240 
2241 	/*
2242 	 * Create the pool's history object.
2243 	 */
2244 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2245 		spa_history_create_obj(spa, tx);
2246 
2247 	/*
2248 	 * Set pool properties.
2249 	 */
2250 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2251 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2252 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2253 	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2254 	if (props != NULL) {
2255 		spa_configfile_set(spa, props, B_FALSE);
2256 		spa_sync_props(spa, props, CRED(), tx);
2257 	}
2258 
2259 	dmu_tx_commit(tx);
2260 
2261 	spa->spa_sync_on = B_TRUE;
2262 	txg_sync_start(spa->spa_dsl_pool);
2263 
2264 	/*
2265 	 * We explicitly wait for the first transaction to complete so that our
2266 	 * bean counters are appropriately updated.
2267 	 */
2268 	txg_wait_synced(spa->spa_dsl_pool, txg);
2269 
2270 	spa_config_sync(spa, B_FALSE, B_TRUE);
2271 
2272 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2273 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2274 	spa_history_log_version(spa, LOG_POOL_CREATE);
2275 
2276 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2277 
2278 	mutex_exit(&spa_namespace_lock);
2279 
2280 	return (0);
2281 }
2282 
2283 #ifdef _KERNEL
2284 /*
2285  * Get the root pool information from the root disk, then import the root pool
2286  * during the system boot up time.
2287  */
2288 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2289 
2290 static nvlist_t *
2291 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2292 {
2293 	nvlist_t *config;
2294 	nvlist_t *nvtop, *nvroot;
2295 	uint64_t pgid;
2296 
2297 	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2298 		return (NULL);
2299 
2300 	/*
2301 	 * Add this top-level vdev to the child array.
2302 	 */
2303 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2304 	    &nvtop) == 0);
2305 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2306 	    &pgid) == 0);
2307 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2308 
2309 	/*
2310 	 * Put this pool's top-level vdevs into a root vdev.
2311 	 */
2312 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2313 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2314 	    VDEV_TYPE_ROOT) == 0);
2315 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2316 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2317 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2318 	    &nvtop, 1) == 0);
2319 
2320 	/*
2321 	 * Replace the existing vdev_tree with the new root vdev in
2322 	 * this pool's configuration (remove the old, add the new).
2323 	 */
2324 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2325 	nvlist_free(nvroot);
2326 	return (config);
2327 }
2328 
2329 /*
2330  * Walk the vdev tree and see if we can find a device with "better"
2331  * configuration. A configuration is "better" if the label on that
2332  * device has a more recent txg.
2333  */
2334 static void
2335 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2336 {
2337 	for (int c = 0; c < vd->vdev_children; c++)
2338 		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2339 
2340 	if (vd->vdev_ops->vdev_op_leaf) {
2341 		nvlist_t *label;
2342 		uint64_t label_txg;
2343 
2344 		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2345 		    &label) != 0)
2346 			return;
2347 
2348 		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2349 		    &label_txg) == 0);
2350 
2351 		/*
2352 		 * Do we have a better boot device?
2353 		 */
2354 		if (label_txg > *txg) {
2355 			*txg = label_txg;
2356 			*avd = vd;
2357 		}
2358 		nvlist_free(label);
2359 	}
2360 }
2361 
2362 /*
2363  * Import a root pool.
2364  *
2365  * For x86. devpath_list will consist of devid and/or physpath name of
2366  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2367  * The GRUB "findroot" command will return the vdev we should boot.
2368  *
2369  * For Sparc, devpath_list consists the physpath name of the booting device
2370  * no matter the rootpool is a single device pool or a mirrored pool.
2371  * e.g.
2372  *	"/pci@1f,0/ide@d/disk@0,0:a"
2373  */
2374 int
2375 spa_import_rootpool(char *devpath, char *devid)
2376 {
2377 	spa_t *spa;
2378 	vdev_t *rvd, *bvd, *avd = NULL;
2379 	nvlist_t *config, *nvtop;
2380 	uint64_t guid, txg;
2381 	char *pname;
2382 	int error;
2383 
2384 	/*
2385 	 * Read the label from the boot device and generate a configuration.
2386 	 */
2387 	if ((config = spa_generate_rootconf(devpath, devid, &guid)) == NULL) {
2388 		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
2389 		    devpath);
2390 		return (EIO);
2391 	}
2392 
2393 	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
2394 	    &pname) == 0);
2395 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2396 
2397 	mutex_enter(&spa_namespace_lock);
2398 	if ((spa = spa_lookup(pname)) != NULL) {
2399 		/*
2400 		 * Remove the existing root pool from the namespace so that we
2401 		 * can replace it with the correct config we just read in.
2402 		 */
2403 		spa_remove(spa);
2404 	}
2405 
2406 	spa = spa_add(pname, NULL);
2407 	spa->spa_is_root = B_TRUE;
2408 
2409 	/*
2410 	 * Build up a vdev tree based on the boot device's label config.
2411 	 */
2412 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2413 	    &nvtop) == 0);
2414 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2415 	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
2416 	    VDEV_ALLOC_ROOTPOOL);
2417 	spa_config_exit(spa, SCL_ALL, FTAG);
2418 	if (error) {
2419 		mutex_exit(&spa_namespace_lock);
2420 		nvlist_free(config);
2421 		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
2422 		    pname);
2423 		return (error);
2424 	}
2425 
2426 	/*
2427 	 * Get the boot vdev.
2428 	 */
2429 	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2430 		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
2431 		    (u_longlong_t)guid);
2432 		error = ENOENT;
2433 		goto out;
2434 	}
2435 
2436 	/*
2437 	 * Determine if there is a better boot device.
2438 	 */
2439 	avd = bvd;
2440 	spa_alt_rootvdev(rvd, &avd, &txg);
2441 	if (avd != bvd) {
2442 		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
2443 		    "try booting from '%s'", avd->vdev_path);
2444 		error = EINVAL;
2445 		goto out;
2446 	}
2447 
2448 	/*
2449 	 * If the boot device is part of a spare vdev then ensure that
2450 	 * we're booting off the active spare.
2451 	 */
2452 	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2453 	    !bvd->vdev_isspare) {
2454 		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
2455 		    "try booting from '%s'",
2456 		    bvd->vdev_parent->vdev_child[1]->vdev_path);
2457 		error = EINVAL;
2458 		goto out;
2459 	}
2460 
2461 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2462 	error = 0;
2463 	spa_history_log_version(spa, LOG_POOL_IMPORT);
2464 out:
2465 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2466 	vdev_free(rvd);
2467 	spa_config_exit(spa, SCL_ALL, FTAG);
2468 	mutex_exit(&spa_namespace_lock);
2469 
2470 	nvlist_free(config);
2471 	return (error);
2472 }
2473 
2474 #endif
2475 
2476 /*
2477  * Take a pool and insert it into the namespace as if it had been loaded at
2478  * boot.
2479  */
2480 int
2481 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2482 {
2483 	spa_t *spa;
2484 	char *altroot = NULL;
2485 
2486 	mutex_enter(&spa_namespace_lock);
2487 	if (spa_lookup(pool) != NULL) {
2488 		mutex_exit(&spa_namespace_lock);
2489 		return (EEXIST);
2490 	}
2491 
2492 	(void) nvlist_lookup_string(props,
2493 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2494 	spa = spa_add(pool, altroot);
2495 
2496 	spa->spa_inactive_states_ok = B_TRUE;
2497 
2498 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2499 
2500 	if (props != NULL)
2501 		spa_configfile_set(spa, props, B_FALSE);
2502 
2503 	spa_config_sync(spa, B_FALSE, B_TRUE);
2504 
2505 	mutex_exit(&spa_namespace_lock);
2506 	spa_history_log_version(spa, LOG_POOL_IMPORT);
2507 
2508 	return (0);
2509 }
2510 
2511 /*
2512  * Import a non-root pool into the system.
2513  */
2514 int
2515 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2516 {
2517 	spa_t *spa;
2518 	char *altroot = NULL;
2519 	int error;
2520 	nvlist_t *nvroot;
2521 	nvlist_t **spares, **l2cache;
2522 	uint_t nspares, nl2cache;
2523 
2524 	/*
2525 	 * If a pool with this name exists, return failure.
2526 	 */
2527 	mutex_enter(&spa_namespace_lock);
2528 	if ((spa = spa_lookup(pool)) != NULL) {
2529 		mutex_exit(&spa_namespace_lock);
2530 		return (EEXIST);
2531 	}
2532 
2533 	/*
2534 	 * Create and initialize the spa structure.
2535 	 */
2536 	(void) nvlist_lookup_string(props,
2537 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2538 	spa = spa_add(pool, altroot);
2539 	spa_activate(spa, spa_mode_global);
2540 
2541 	/*
2542 	 * Don't start async tasks until we know everything is healthy.
2543 	 */
2544 	spa_async_suspend(spa);
2545 
2546 	/*
2547 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2548 	 * because the user-supplied config is actually the one to trust when
2549 	 * doing an import.
2550 	 */
2551 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
2552 
2553 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2554 	/*
2555 	 * Toss any existing sparelist, as it doesn't have any validity
2556 	 * anymore, and conflicts with spa_has_spare().
2557 	 */
2558 	if (spa->spa_spares.sav_config) {
2559 		nvlist_free(spa->spa_spares.sav_config);
2560 		spa->spa_spares.sav_config = NULL;
2561 		spa_load_spares(spa);
2562 	}
2563 	if (spa->spa_l2cache.sav_config) {
2564 		nvlist_free(spa->spa_l2cache.sav_config);
2565 		spa->spa_l2cache.sav_config = NULL;
2566 		spa_load_l2cache(spa);
2567 	}
2568 
2569 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2570 	    &nvroot) == 0);
2571 	if (error == 0)
2572 		error = spa_validate_aux(spa, nvroot, -1ULL,
2573 		    VDEV_ALLOC_SPARE);
2574 	if (error == 0)
2575 		error = spa_validate_aux(spa, nvroot, -1ULL,
2576 		    VDEV_ALLOC_L2CACHE);
2577 	spa_config_exit(spa, SCL_ALL, FTAG);
2578 
2579 	if (props != NULL)
2580 		spa_configfile_set(spa, props, B_FALSE);
2581 
2582 	if (error != 0 || (props && spa_writeable(spa) &&
2583 	    (error = spa_prop_set(spa, props)))) {
2584 		spa_unload(spa);
2585 		spa_deactivate(spa);
2586 		spa_remove(spa);
2587 		mutex_exit(&spa_namespace_lock);
2588 		return (error);
2589 	}
2590 
2591 	spa_async_resume(spa);
2592 
2593 	/*
2594 	 * Override any spares and level 2 cache devices as specified by
2595 	 * the user, as these may have correct device names/devids, etc.
2596 	 */
2597 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2598 	    &spares, &nspares) == 0) {
2599 		if (spa->spa_spares.sav_config)
2600 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2601 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2602 		else
2603 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2604 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2605 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2606 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2607 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2608 		spa_load_spares(spa);
2609 		spa_config_exit(spa, SCL_ALL, FTAG);
2610 		spa->spa_spares.sav_sync = B_TRUE;
2611 	}
2612 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2613 	    &l2cache, &nl2cache) == 0) {
2614 		if (spa->spa_l2cache.sav_config)
2615 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2616 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2617 		else
2618 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2619 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2620 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2621 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2622 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2623 		spa_load_l2cache(spa);
2624 		spa_config_exit(spa, SCL_ALL, FTAG);
2625 		spa->spa_l2cache.sav_sync = B_TRUE;
2626 	}
2627 
2628 	if (spa_writeable(spa)) {
2629 		/*
2630 		 * Update the config cache to include the newly-imported pool.
2631 		 */
2632 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
2633 	}
2634 
2635 	/*
2636 	 * It's possible that the pool was expanded while it was exported.
2637 	 * We kick off an async task to handle this for us.
2638 	 */
2639 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
2640 
2641 	mutex_exit(&spa_namespace_lock);
2642 	spa_history_log_version(spa, LOG_POOL_IMPORT);
2643 
2644 	return (0);
2645 }
2646 
2647 
2648 /*
2649  * This (illegal) pool name is used when temporarily importing a spa_t in order
2650  * to get the vdev stats associated with the imported devices.
2651  */
2652 #define	TRYIMPORT_NAME	"$import"
2653 
2654 nvlist_t *
2655 spa_tryimport(nvlist_t *tryconfig)
2656 {
2657 	nvlist_t *config = NULL;
2658 	char *poolname;
2659 	spa_t *spa;
2660 	uint64_t state;
2661 	int error;
2662 
2663 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2664 		return (NULL);
2665 
2666 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2667 		return (NULL);
2668 
2669 	/*
2670 	 * Create and initialize the spa structure.
2671 	 */
2672 	mutex_enter(&spa_namespace_lock);
2673 	spa = spa_add(TRYIMPORT_NAME, NULL);
2674 	spa_activate(spa, FREAD);
2675 
2676 	/*
2677 	 * Pass off the heavy lifting to spa_load().
2678 	 * Pass TRUE for mosconfig because the user-supplied config
2679 	 * is actually the one to trust when doing an import.
2680 	 */
2681 	error = spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2682 
2683 	/*
2684 	 * If 'tryconfig' was at least parsable, return the current config.
2685 	 */
2686 	if (spa->spa_root_vdev != NULL) {
2687 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2688 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2689 		    poolname) == 0);
2690 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2691 		    state) == 0);
2692 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2693 		    spa->spa_uberblock.ub_timestamp) == 0);
2694 
2695 		/*
2696 		 * If the bootfs property exists on this pool then we
2697 		 * copy it out so that external consumers can tell which
2698 		 * pools are bootable.
2699 		 */
2700 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
2701 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2702 
2703 			/*
2704 			 * We have to play games with the name since the
2705 			 * pool was opened as TRYIMPORT_NAME.
2706 			 */
2707 			if (dsl_dsobj_to_dsname(spa_name(spa),
2708 			    spa->spa_bootfs, tmpname) == 0) {
2709 				char *cp;
2710 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2711 
2712 				cp = strchr(tmpname, '/');
2713 				if (cp == NULL) {
2714 					(void) strlcpy(dsname, tmpname,
2715 					    MAXPATHLEN);
2716 				} else {
2717 					(void) snprintf(dsname, MAXPATHLEN,
2718 					    "%s/%s", poolname, ++cp);
2719 				}
2720 				VERIFY(nvlist_add_string(config,
2721 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2722 				kmem_free(dsname, MAXPATHLEN);
2723 			}
2724 			kmem_free(tmpname, MAXPATHLEN);
2725 		}
2726 
2727 		/*
2728 		 * Add the list of hot spares and level 2 cache devices.
2729 		 */
2730 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2731 		spa_add_spares(spa, config);
2732 		spa_add_l2cache(spa, config);
2733 		spa_config_exit(spa, SCL_CONFIG, FTAG);
2734 	}
2735 
2736 	spa_unload(spa);
2737 	spa_deactivate(spa);
2738 	spa_remove(spa);
2739 	mutex_exit(&spa_namespace_lock);
2740 
2741 	return (config);
2742 }
2743 
2744 /*
2745  * Pool export/destroy
2746  *
2747  * The act of destroying or exporting a pool is very simple.  We make sure there
2748  * is no more pending I/O and any references to the pool are gone.  Then, we
2749  * update the pool state and sync all the labels to disk, removing the
2750  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2751  * we don't sync the labels or remove the configuration cache.
2752  */
2753 static int
2754 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2755     boolean_t force, boolean_t hardforce)
2756 {
2757 	spa_t *spa;
2758 
2759 	if (oldconfig)
2760 		*oldconfig = NULL;
2761 
2762 	if (!(spa_mode_global & FWRITE))
2763 		return (EROFS);
2764 
2765 	mutex_enter(&spa_namespace_lock);
2766 	if ((spa = spa_lookup(pool)) == NULL) {
2767 		mutex_exit(&spa_namespace_lock);
2768 		return (ENOENT);
2769 	}
2770 
2771 	/*
2772 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2773 	 * reacquire the namespace lock, and see if we can export.
2774 	 */
2775 	spa_open_ref(spa, FTAG);
2776 	mutex_exit(&spa_namespace_lock);
2777 	spa_async_suspend(spa);
2778 	mutex_enter(&spa_namespace_lock);
2779 	spa_close(spa, FTAG);
2780 
2781 	/*
2782 	 * The pool will be in core if it's openable,
2783 	 * in which case we can modify its state.
2784 	 */
2785 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2786 		/*
2787 		 * Objsets may be open only because they're dirty, so we
2788 		 * have to force it to sync before checking spa_refcnt.
2789 		 */
2790 		txg_wait_synced(spa->spa_dsl_pool, 0);
2791 
2792 		/*
2793 		 * A pool cannot be exported or destroyed if there are active
2794 		 * references.  If we are resetting a pool, allow references by
2795 		 * fault injection handlers.
2796 		 */
2797 		if (!spa_refcount_zero(spa) ||
2798 		    (spa->spa_inject_ref != 0 &&
2799 		    new_state != POOL_STATE_UNINITIALIZED)) {
2800 			spa_async_resume(spa);
2801 			mutex_exit(&spa_namespace_lock);
2802 			return (EBUSY);
2803 		}
2804 
2805 		/*
2806 		 * A pool cannot be exported if it has an active shared spare.
2807 		 * This is to prevent other pools stealing the active spare
2808 		 * from an exported pool. At user's own will, such pool can
2809 		 * be forcedly exported.
2810 		 */
2811 		if (!force && new_state == POOL_STATE_EXPORTED &&
2812 		    spa_has_active_shared_spare(spa)) {
2813 			spa_async_resume(spa);
2814 			mutex_exit(&spa_namespace_lock);
2815 			return (EXDEV);
2816 		}
2817 
2818 		/*
2819 		 * We want this to be reflected on every label,
2820 		 * so mark them all dirty.  spa_unload() will do the
2821 		 * final sync that pushes these changes out.
2822 		 */
2823 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
2824 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2825 			spa->spa_state = new_state;
2826 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2827 			vdev_config_dirty(spa->spa_root_vdev);
2828 			spa_config_exit(spa, SCL_ALL, FTAG);
2829 		}
2830 	}
2831 
2832 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2833 
2834 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2835 		spa_unload(spa);
2836 		spa_deactivate(spa);
2837 	}
2838 
2839 	if (oldconfig && spa->spa_config)
2840 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2841 
2842 	if (new_state != POOL_STATE_UNINITIALIZED) {
2843 		if (!hardforce)
2844 			spa_config_sync(spa, B_TRUE, B_TRUE);
2845 		spa_remove(spa);
2846 	}
2847 	mutex_exit(&spa_namespace_lock);
2848 
2849 	return (0);
2850 }
2851 
2852 /*
2853  * Destroy a storage pool.
2854  */
2855 int
2856 spa_destroy(char *pool)
2857 {
2858 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
2859 	    B_FALSE, B_FALSE));
2860 }
2861 
2862 /*
2863  * Export a storage pool.
2864  */
2865 int
2866 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
2867     boolean_t hardforce)
2868 {
2869 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
2870 	    force, hardforce));
2871 }
2872 
2873 /*
2874  * Similar to spa_export(), this unloads the spa_t without actually removing it
2875  * from the namespace in any way.
2876  */
2877 int
2878 spa_reset(char *pool)
2879 {
2880 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2881 	    B_FALSE, B_FALSE));
2882 }
2883 
2884 /*
2885  * ==========================================================================
2886  * Device manipulation
2887  * ==========================================================================
2888  */
2889 
2890 /*
2891  * Add a device to a storage pool.
2892  */
2893 int
2894 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2895 {
2896 	uint64_t txg;
2897 	int error;
2898 	vdev_t *rvd = spa->spa_root_vdev;
2899 	vdev_t *vd, *tvd;
2900 	nvlist_t **spares, **l2cache;
2901 	uint_t nspares, nl2cache;
2902 
2903 	txg = spa_vdev_enter(spa);
2904 
2905 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2906 	    VDEV_ALLOC_ADD)) != 0)
2907 		return (spa_vdev_exit(spa, NULL, txg, error));
2908 
2909 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
2910 
2911 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2912 	    &nspares) != 0)
2913 		nspares = 0;
2914 
2915 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2916 	    &nl2cache) != 0)
2917 		nl2cache = 0;
2918 
2919 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2920 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2921 
2922 	if (vd->vdev_children != 0 &&
2923 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
2924 		return (spa_vdev_exit(spa, vd, txg, error));
2925 
2926 	/*
2927 	 * We must validate the spares and l2cache devices after checking the
2928 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2929 	 */
2930 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2931 		return (spa_vdev_exit(spa, vd, txg, error));
2932 
2933 	/*
2934 	 * Transfer each new top-level vdev from vd to rvd.
2935 	 */
2936 	for (int c = 0; c < vd->vdev_children; c++) {
2937 		tvd = vd->vdev_child[c];
2938 		vdev_remove_child(vd, tvd);
2939 		tvd->vdev_id = rvd->vdev_children;
2940 		vdev_add_child(rvd, tvd);
2941 		vdev_config_dirty(tvd);
2942 	}
2943 
2944 	if (nspares != 0) {
2945 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2946 		    ZPOOL_CONFIG_SPARES);
2947 		spa_load_spares(spa);
2948 		spa->spa_spares.sav_sync = B_TRUE;
2949 	}
2950 
2951 	if (nl2cache != 0) {
2952 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2953 		    ZPOOL_CONFIG_L2CACHE);
2954 		spa_load_l2cache(spa);
2955 		spa->spa_l2cache.sav_sync = B_TRUE;
2956 	}
2957 
2958 	/*
2959 	 * We have to be careful when adding new vdevs to an existing pool.
2960 	 * If other threads start allocating from these vdevs before we
2961 	 * sync the config cache, and we lose power, then upon reboot we may
2962 	 * fail to open the pool because there are DVAs that the config cache
2963 	 * can't translate.  Therefore, we first add the vdevs without
2964 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2965 	 * and then let spa_config_update() initialize the new metaslabs.
2966 	 *
2967 	 * spa_load() checks for added-but-not-initialized vdevs, so that
2968 	 * if we lose power at any point in this sequence, the remaining
2969 	 * steps will be completed the next time we load the pool.
2970 	 */
2971 	(void) spa_vdev_exit(spa, vd, txg, 0);
2972 
2973 	mutex_enter(&spa_namespace_lock);
2974 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2975 	mutex_exit(&spa_namespace_lock);
2976 
2977 	return (0);
2978 }
2979 
2980 /*
2981  * Attach a device to a mirror.  The arguments are the path to any device
2982  * in the mirror, and the nvroot for the new device.  If the path specifies
2983  * a device that is not mirrored, we automatically insert the mirror vdev.
2984  *
2985  * If 'replacing' is specified, the new device is intended to replace the
2986  * existing device; in this case the two devices are made into their own
2987  * mirror using the 'replacing' vdev, which is functionally identical to
2988  * the mirror vdev (it actually reuses all the same ops) but has a few
2989  * extra rules: you can't attach to it after it's been created, and upon
2990  * completion of resilvering, the first disk (the one being replaced)
2991  * is automatically detached.
2992  */
2993 int
2994 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2995 {
2996 	uint64_t txg, open_txg;
2997 	vdev_t *rvd = spa->spa_root_vdev;
2998 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2999 	vdev_ops_t *pvops;
3000 	char *oldvdpath, *newvdpath;
3001 	int newvd_isspare;
3002 	int error;
3003 
3004 	txg = spa_vdev_enter(spa);
3005 
3006 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3007 
3008 	if (oldvd == NULL)
3009 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3010 
3011 	if (!oldvd->vdev_ops->vdev_op_leaf)
3012 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3013 
3014 	pvd = oldvd->vdev_parent;
3015 
3016 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3017 	    VDEV_ALLOC_ADD)) != 0)
3018 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3019 
3020 	if (newrootvd->vdev_children != 1)
3021 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3022 
3023 	newvd = newrootvd->vdev_child[0];
3024 
3025 	if (!newvd->vdev_ops->vdev_op_leaf)
3026 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3027 
3028 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3029 		return (spa_vdev_exit(spa, newrootvd, txg, error));
3030 
3031 	/*
3032 	 * Spares can't replace logs
3033 	 */
3034 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3035 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3036 
3037 	if (!replacing) {
3038 		/*
3039 		 * For attach, the only allowable parent is a mirror or the root
3040 		 * vdev.
3041 		 */
3042 		if (pvd->vdev_ops != &vdev_mirror_ops &&
3043 		    pvd->vdev_ops != &vdev_root_ops)
3044 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3045 
3046 		pvops = &vdev_mirror_ops;
3047 	} else {
3048 		/*
3049 		 * Active hot spares can only be replaced by inactive hot
3050 		 * spares.
3051 		 */
3052 		if (pvd->vdev_ops == &vdev_spare_ops &&
3053 		    pvd->vdev_child[1] == oldvd &&
3054 		    !spa_has_spare(spa, newvd->vdev_guid))
3055 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3056 
3057 		/*
3058 		 * If the source is a hot spare, and the parent isn't already a
3059 		 * spare, then we want to create a new hot spare.  Otherwise, we
3060 		 * want to create a replacing vdev.  The user is not allowed to
3061 		 * attach to a spared vdev child unless the 'isspare' state is
3062 		 * the same (spare replaces spare, non-spare replaces
3063 		 * non-spare).
3064 		 */
3065 		if (pvd->vdev_ops == &vdev_replacing_ops)
3066 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3067 		else if (pvd->vdev_ops == &vdev_spare_ops &&
3068 		    newvd->vdev_isspare != oldvd->vdev_isspare)
3069 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3070 		else if (pvd->vdev_ops != &vdev_spare_ops &&
3071 		    newvd->vdev_isspare)
3072 			pvops = &vdev_spare_ops;
3073 		else
3074 			pvops = &vdev_replacing_ops;
3075 	}
3076 
3077 	/*
3078 	 * Make sure the new device is big enough.
3079 	 */
3080 	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3081 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3082 
3083 	/*
3084 	 * The new device cannot have a higher alignment requirement
3085 	 * than the top-level vdev.
3086 	 */
3087 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3088 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3089 
3090 	/*
3091 	 * If this is an in-place replacement, update oldvd's path and devid
3092 	 * to make it distinguishable from newvd, and unopenable from now on.
3093 	 */
3094 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3095 		spa_strfree(oldvd->vdev_path);
3096 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3097 		    KM_SLEEP);
3098 		(void) sprintf(oldvd->vdev_path, "%s/%s",
3099 		    newvd->vdev_path, "old");
3100 		if (oldvd->vdev_devid != NULL) {
3101 			spa_strfree(oldvd->vdev_devid);
3102 			oldvd->vdev_devid = NULL;
3103 		}
3104 	}
3105 
3106 	/*
3107 	 * If the parent is not a mirror, or if we're replacing, insert the new
3108 	 * mirror/replacing/spare vdev above oldvd.
3109 	 */
3110 	if (pvd->vdev_ops != pvops)
3111 		pvd = vdev_add_parent(oldvd, pvops);
3112 
3113 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3114 	ASSERT(pvd->vdev_ops == pvops);
3115 	ASSERT(oldvd->vdev_parent == pvd);
3116 
3117 	/*
3118 	 * Extract the new device from its root and add it to pvd.
3119 	 */
3120 	vdev_remove_child(newrootvd, newvd);
3121 	newvd->vdev_id = pvd->vdev_children;
3122 	vdev_add_child(pvd, newvd);
3123 
3124 	tvd = newvd->vdev_top;
3125 	ASSERT(pvd->vdev_top == tvd);
3126 	ASSERT(tvd->vdev_parent == rvd);
3127 
3128 	vdev_config_dirty(tvd);
3129 
3130 	/*
3131 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3132 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3133 	 */
3134 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3135 
3136 	vdev_dtl_dirty(newvd, DTL_MISSING,
3137 	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3138 
3139 	if (newvd->vdev_isspare) {
3140 		spa_spare_activate(newvd);
3141 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3142 	}
3143 
3144 	oldvdpath = spa_strdup(oldvd->vdev_path);
3145 	newvdpath = spa_strdup(newvd->vdev_path);
3146 	newvd_isspare = newvd->vdev_isspare;
3147 
3148 	/*
3149 	 * Mark newvd's DTL dirty in this txg.
3150 	 */
3151 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3152 
3153 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3154 
3155 	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3156 	    CRED(),  "%s vdev=%s %s vdev=%s",
3157 	    replacing && newvd_isspare ? "spare in" :
3158 	    replacing ? "replace" : "attach", newvdpath,
3159 	    replacing ? "for" : "to", oldvdpath);
3160 
3161 	spa_strfree(oldvdpath);
3162 	spa_strfree(newvdpath);
3163 
3164 	/*
3165 	 * Kick off a resilver to update newvd.
3166 	 */
3167 	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3168 
3169 	return (0);
3170 }
3171 
3172 /*
3173  * Detach a device from a mirror or replacing vdev.
3174  * If 'replace_done' is specified, only detach if the parent
3175  * is a replacing vdev.
3176  */
3177 int
3178 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3179 {
3180 	uint64_t txg;
3181 	int error;
3182 	vdev_t *rvd = spa->spa_root_vdev;
3183 	vdev_t *vd, *pvd, *cvd, *tvd;
3184 	boolean_t unspare = B_FALSE;
3185 	uint64_t unspare_guid;
3186 	size_t len;
3187 
3188 	txg = spa_vdev_enter(spa);
3189 
3190 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3191 
3192 	if (vd == NULL)
3193 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3194 
3195 	if (!vd->vdev_ops->vdev_op_leaf)
3196 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3197 
3198 	pvd = vd->vdev_parent;
3199 
3200 	/*
3201 	 * If the parent/child relationship is not as expected, don't do it.
3202 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3203 	 * vdev that's replacing B with C.  The user's intent in replacing
3204 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3205 	 * the replace by detaching C, the expected behavior is to end up
3206 	 * M(A,B).  But suppose that right after deciding to detach C,
3207 	 * the replacement of B completes.  We would have M(A,C), and then
3208 	 * ask to detach C, which would leave us with just A -- not what
3209 	 * the user wanted.  To prevent this, we make sure that the
3210 	 * parent/child relationship hasn't changed -- in this example,
3211 	 * that C's parent is still the replacing vdev R.
3212 	 */
3213 	if (pvd->vdev_guid != pguid && pguid != 0)
3214 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3215 
3216 	/*
3217 	 * If replace_done is specified, only remove this device if it's
3218 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3219 	 * disk can be removed.
3220 	 */
3221 	if (replace_done) {
3222 		if (pvd->vdev_ops == &vdev_replacing_ops) {
3223 			if (vd->vdev_id != 0)
3224 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3225 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3226 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3227 		}
3228 	}
3229 
3230 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3231 	    spa_version(spa) >= SPA_VERSION_SPARES);
3232 
3233 	/*
3234 	 * Only mirror, replacing, and spare vdevs support detach.
3235 	 */
3236 	if (pvd->vdev_ops != &vdev_replacing_ops &&
3237 	    pvd->vdev_ops != &vdev_mirror_ops &&
3238 	    pvd->vdev_ops != &vdev_spare_ops)
3239 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3240 
3241 	/*
3242 	 * If this device has the only valid copy of some data,
3243 	 * we cannot safely detach it.
3244 	 */
3245 	if (vdev_dtl_required(vd))
3246 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3247 
3248 	ASSERT(pvd->vdev_children >= 2);
3249 
3250 	/*
3251 	 * If we are detaching the second disk from a replacing vdev, then
3252 	 * check to see if we changed the original vdev's path to have "/old"
3253 	 * at the end in spa_vdev_attach().  If so, undo that change now.
3254 	 */
3255 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3256 	    pvd->vdev_child[0]->vdev_path != NULL &&
3257 	    pvd->vdev_child[1]->vdev_path != NULL) {
3258 		ASSERT(pvd->vdev_child[1] == vd);
3259 		cvd = pvd->vdev_child[0];
3260 		len = strlen(vd->vdev_path);
3261 		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3262 		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3263 			spa_strfree(cvd->vdev_path);
3264 			cvd->vdev_path = spa_strdup(vd->vdev_path);
3265 		}
3266 	}
3267 
3268 	/*
3269 	 * If we are detaching the original disk from a spare, then it implies
3270 	 * that the spare should become a real disk, and be removed from the
3271 	 * active spare list for the pool.
3272 	 */
3273 	if (pvd->vdev_ops == &vdev_spare_ops &&
3274 	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3275 		unspare = B_TRUE;
3276 
3277 	/*
3278 	 * Erase the disk labels so the disk can be used for other things.
3279 	 * This must be done after all other error cases are handled,
3280 	 * but before we disembowel vd (so we can still do I/O to it).
3281 	 * But if we can't do it, don't treat the error as fatal --
3282 	 * it may be that the unwritability of the disk is the reason
3283 	 * it's being detached!
3284 	 */
3285 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3286 
3287 	/*
3288 	 * Remove vd from its parent and compact the parent's children.
3289 	 */
3290 	vdev_remove_child(pvd, vd);
3291 	vdev_compact_children(pvd);
3292 
3293 	/*
3294 	 * Remember one of the remaining children so we can get tvd below.
3295 	 */
3296 	cvd = pvd->vdev_child[0];
3297 
3298 	/*
3299 	 * If we need to remove the remaining child from the list of hot spares,
3300 	 * do it now, marking the vdev as no longer a spare in the process.
3301 	 * We must do this before vdev_remove_parent(), because that can
3302 	 * change the GUID if it creates a new toplevel GUID.  For a similar
3303 	 * reason, we must remove the spare now, in the same txg as the detach;
3304 	 * otherwise someone could attach a new sibling, change the GUID, and
3305 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3306 	 */
3307 	if (unspare) {
3308 		ASSERT(cvd->vdev_isspare);
3309 		spa_spare_remove(cvd);
3310 		unspare_guid = cvd->vdev_guid;
3311 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3312 	}
3313 
3314 	/*
3315 	 * If the parent mirror/replacing vdev only has one child,
3316 	 * the parent is no longer needed.  Remove it from the tree.
3317 	 */
3318 	if (pvd->vdev_children == 1)
3319 		vdev_remove_parent(cvd);
3320 
3321 	/*
3322 	 * We don't set tvd until now because the parent we just removed
3323 	 * may have been the previous top-level vdev.
3324 	 */
3325 	tvd = cvd->vdev_top;
3326 	ASSERT(tvd->vdev_parent == rvd);
3327 
3328 	/*
3329 	 * Reevaluate the parent vdev state.
3330 	 */
3331 	vdev_propagate_state(cvd);
3332 
3333 	/*
3334 	 * If the 'autoexpand' property is set on the pool then automatically
3335 	 * try to expand the size of the pool. For example if the device we
3336 	 * just detached was smaller than the others, it may be possible to
3337 	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3338 	 * first so that we can obtain the updated sizes of the leaf vdevs.
3339 	 */
3340 	if (spa->spa_autoexpand) {
3341 		vdev_reopen(tvd);
3342 		vdev_expand(tvd, txg);
3343 	}
3344 
3345 	vdev_config_dirty(tvd);
3346 
3347 	/*
3348 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3349 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3350 	 * But first make sure we're not on any *other* txg's DTL list, to
3351 	 * prevent vd from being accessed after it's freed.
3352 	 */
3353 	for (int t = 0; t < TXG_SIZE; t++)
3354 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3355 	vd->vdev_detached = B_TRUE;
3356 	vdev_dirty(tvd, VDD_DTL, vd, txg);
3357 
3358 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3359 
3360 	error = spa_vdev_exit(spa, vd, txg, 0);
3361 
3362 	/*
3363 	 * If this was the removal of the original device in a hot spare vdev,
3364 	 * then we want to go through and remove the device from the hot spare
3365 	 * list of every other pool.
3366 	 */
3367 	if (unspare) {
3368 		spa_t *myspa = spa;
3369 		spa = NULL;
3370 		mutex_enter(&spa_namespace_lock);
3371 		while ((spa = spa_next(spa)) != NULL) {
3372 			if (spa->spa_state != POOL_STATE_ACTIVE)
3373 				continue;
3374 			if (spa == myspa)
3375 				continue;
3376 			spa_open_ref(spa, FTAG);
3377 			mutex_exit(&spa_namespace_lock);
3378 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3379 			mutex_enter(&spa_namespace_lock);
3380 			spa_close(spa, FTAG);
3381 		}
3382 		mutex_exit(&spa_namespace_lock);
3383 	}
3384 
3385 	return (error);
3386 }
3387 
3388 static nvlist_t *
3389 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3390 {
3391 	for (int i = 0; i < count; i++) {
3392 		uint64_t guid;
3393 
3394 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3395 		    &guid) == 0);
3396 
3397 		if (guid == target_guid)
3398 			return (nvpp[i]);
3399 	}
3400 
3401 	return (NULL);
3402 }
3403 
3404 static void
3405 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3406 	nvlist_t *dev_to_remove)
3407 {
3408 	nvlist_t **newdev = NULL;
3409 
3410 	if (count > 1)
3411 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3412 
3413 	for (int i = 0, j = 0; i < count; i++) {
3414 		if (dev[i] == dev_to_remove)
3415 			continue;
3416 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3417 	}
3418 
3419 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3420 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3421 
3422 	for (int i = 0; i < count - 1; i++)
3423 		nvlist_free(newdev[i]);
3424 
3425 	if (count > 1)
3426 		kmem_free(newdev, (count - 1) * sizeof (void *));
3427 }
3428 
3429 /*
3430  * Remove a device from the pool.  Currently, this supports removing only hot
3431  * spares and level 2 ARC devices.
3432  */
3433 int
3434 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3435 {
3436 	vdev_t *vd;
3437 	nvlist_t **spares, **l2cache, *nv;
3438 	uint_t nspares, nl2cache;
3439 	uint64_t txg = 0;
3440 	int error = 0;
3441 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3442 
3443 	if (!locked)
3444 		txg = spa_vdev_enter(spa);
3445 
3446 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3447 
3448 	if (spa->spa_spares.sav_vdevs != NULL &&
3449 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3450 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3451 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3452 		/*
3453 		 * Only remove the hot spare if it's not currently in use
3454 		 * in this pool.
3455 		 */
3456 		if (vd == NULL || unspare) {
3457 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3458 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3459 			spa_load_spares(spa);
3460 			spa->spa_spares.sav_sync = B_TRUE;
3461 		} else {
3462 			error = EBUSY;
3463 		}
3464 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3465 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3466 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3467 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3468 		/*
3469 		 * Cache devices can always be removed.
3470 		 */
3471 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3472 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3473 		spa_load_l2cache(spa);
3474 		spa->spa_l2cache.sav_sync = B_TRUE;
3475 	} else if (vd != NULL) {
3476 		/*
3477 		 * Normal vdevs cannot be removed (yet).
3478 		 */
3479 		error = ENOTSUP;
3480 	} else {
3481 		/*
3482 		 * There is no vdev of any kind with the specified guid.
3483 		 */
3484 		error = ENOENT;
3485 	}
3486 
3487 	if (!locked)
3488 		return (spa_vdev_exit(spa, NULL, txg, error));
3489 
3490 	return (error);
3491 }
3492 
3493 /*
3494  * Find any device that's done replacing, or a vdev marked 'unspare' that's
3495  * current spared, so we can detach it.
3496  */
3497 static vdev_t *
3498 spa_vdev_resilver_done_hunt(vdev_t *vd)
3499 {
3500 	vdev_t *newvd, *oldvd;
3501 
3502 	for (int c = 0; c < vd->vdev_children; c++) {
3503 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3504 		if (oldvd != NULL)
3505 			return (oldvd);
3506 	}
3507 
3508 	/*
3509 	 * Check for a completed replacement.
3510 	 */
3511 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3512 		oldvd = vd->vdev_child[0];
3513 		newvd = vd->vdev_child[1];
3514 
3515 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
3516 		    !vdev_dtl_required(oldvd))
3517 			return (oldvd);
3518 	}
3519 
3520 	/*
3521 	 * Check for a completed resilver with the 'unspare' flag set.
3522 	 */
3523 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3524 		newvd = vd->vdev_child[0];
3525 		oldvd = vd->vdev_child[1];
3526 
3527 		if (newvd->vdev_unspare &&
3528 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
3529 		    !vdev_dtl_required(oldvd)) {
3530 			newvd->vdev_unspare = 0;
3531 			return (oldvd);
3532 		}
3533 	}
3534 
3535 	return (NULL);
3536 }
3537 
3538 static void
3539 spa_vdev_resilver_done(spa_t *spa)
3540 {
3541 	vdev_t *vd, *pvd, *ppvd;
3542 	uint64_t guid, sguid, pguid, ppguid;
3543 
3544 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3545 
3546 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3547 		pvd = vd->vdev_parent;
3548 		ppvd = pvd->vdev_parent;
3549 		guid = vd->vdev_guid;
3550 		pguid = pvd->vdev_guid;
3551 		ppguid = ppvd->vdev_guid;
3552 		sguid = 0;
3553 		/*
3554 		 * If we have just finished replacing a hot spared device, then
3555 		 * we need to detach the parent's first child (the original hot
3556 		 * spare) as well.
3557 		 */
3558 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
3559 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3560 			ASSERT(ppvd->vdev_children == 2);
3561 			sguid = ppvd->vdev_child[1]->vdev_guid;
3562 		}
3563 		spa_config_exit(spa, SCL_ALL, FTAG);
3564 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
3565 			return;
3566 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
3567 			return;
3568 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3569 	}
3570 
3571 	spa_config_exit(spa, SCL_ALL, FTAG);
3572 }
3573 
3574 /*
3575  * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
3576  * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
3577  */
3578 int
3579 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
3580     boolean_t ispath)
3581 {
3582 	vdev_t *vd;
3583 	uint64_t txg;
3584 
3585 	txg = spa_vdev_enter(spa);
3586 
3587 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3588 		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3589 
3590 	if (!vd->vdev_ops->vdev_op_leaf)
3591 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3592 
3593 	if (ispath) {
3594 		spa_strfree(vd->vdev_path);
3595 		vd->vdev_path = spa_strdup(value);
3596 	} else {
3597 		if (vd->vdev_fru != NULL)
3598 			spa_strfree(vd->vdev_fru);
3599 		vd->vdev_fru = spa_strdup(value);
3600 	}
3601 
3602 	vdev_config_dirty(vd->vdev_top);
3603 
3604 	return (spa_vdev_exit(spa, NULL, txg, 0));
3605 }
3606 
3607 int
3608 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3609 {
3610 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
3611 }
3612 
3613 int
3614 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
3615 {
3616 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
3617 }
3618 
3619 /*
3620  * ==========================================================================
3621  * SPA Scrubbing
3622  * ==========================================================================
3623  */
3624 
3625 int
3626 spa_scrub(spa_t *spa, pool_scrub_type_t type)
3627 {
3628 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3629 
3630 	if ((uint_t)type >= POOL_SCRUB_TYPES)
3631 		return (ENOTSUP);
3632 
3633 	/*
3634 	 * If a resilver was requested, but there is no DTL on a
3635 	 * writeable leaf device, we have nothing to do.
3636 	 */
3637 	if (type == POOL_SCRUB_RESILVER &&
3638 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3639 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3640 		return (0);
3641 	}
3642 
3643 	if (type == POOL_SCRUB_EVERYTHING &&
3644 	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3645 	    spa->spa_dsl_pool->dp_scrub_isresilver)
3646 		return (EBUSY);
3647 
3648 	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3649 		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3650 	} else if (type == POOL_SCRUB_NONE) {
3651 		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3652 	} else {
3653 		return (EINVAL);
3654 	}
3655 }
3656 
3657 /*
3658  * ==========================================================================
3659  * SPA async task processing
3660  * ==========================================================================
3661  */
3662 
3663 static void
3664 spa_async_remove(spa_t *spa, vdev_t *vd)
3665 {
3666 	if (vd->vdev_remove_wanted) {
3667 		vd->vdev_remove_wanted = 0;
3668 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3669 		vdev_clear(spa, vd);
3670 		vdev_state_dirty(vd->vdev_top);
3671 	}
3672 
3673 	for (int c = 0; c < vd->vdev_children; c++)
3674 		spa_async_remove(spa, vd->vdev_child[c]);
3675 }
3676 
3677 static void
3678 spa_async_probe(spa_t *spa, vdev_t *vd)
3679 {
3680 	if (vd->vdev_probe_wanted) {
3681 		vd->vdev_probe_wanted = 0;
3682 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
3683 	}
3684 
3685 	for (int c = 0; c < vd->vdev_children; c++)
3686 		spa_async_probe(spa, vd->vdev_child[c]);
3687 }
3688 
3689 static void
3690 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
3691 {
3692 	sysevent_id_t eid;
3693 	nvlist_t *attr;
3694 	char *physpath;
3695 
3696 	if (!spa->spa_autoexpand)
3697 		return;
3698 
3699 	for (int c = 0; c < vd->vdev_children; c++) {
3700 		vdev_t *cvd = vd->vdev_child[c];
3701 		spa_async_autoexpand(spa, cvd);
3702 	}
3703 
3704 	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
3705 		return;
3706 
3707 	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
3708 	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
3709 
3710 	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3711 	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
3712 
3713 	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
3714 	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
3715 
3716 	nvlist_free(attr);
3717 	kmem_free(physpath, MAXPATHLEN);
3718 }
3719 
3720 static void
3721 spa_async_thread(spa_t *spa)
3722 {
3723 	int tasks;
3724 
3725 	ASSERT(spa->spa_sync_on);
3726 
3727 	mutex_enter(&spa->spa_async_lock);
3728 	tasks = spa->spa_async_tasks;
3729 	spa->spa_async_tasks = 0;
3730 	mutex_exit(&spa->spa_async_lock);
3731 
3732 	/*
3733 	 * See if the config needs to be updated.
3734 	 */
3735 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3736 		uint64_t oldsz, space_update;
3737 
3738 		mutex_enter(&spa_namespace_lock);
3739 		oldsz = spa_get_space(spa);
3740 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3741 		space_update = spa_get_space(spa) - oldsz;
3742 		mutex_exit(&spa_namespace_lock);
3743 
3744 		/*
3745 		 * If the pool grew as a result of the config update,
3746 		 * then log an internal history event.
3747 		 */
3748 		if (space_update) {
3749 			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
3750 			    spa, NULL, CRED(),
3751 			    "pool '%s' size: %llu(+%llu)",
3752 			    spa_name(spa), spa_get_space(spa),
3753 			    space_update);
3754 		}
3755 	}
3756 
3757 	/*
3758 	 * See if any devices need to be marked REMOVED.
3759 	 */
3760 	if (tasks & SPA_ASYNC_REMOVE) {
3761 		spa_vdev_state_enter(spa);
3762 		spa_async_remove(spa, spa->spa_root_vdev);
3763 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3764 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3765 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
3766 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3767 		(void) spa_vdev_state_exit(spa, NULL, 0);
3768 	}
3769 
3770 	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
3771 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3772 		spa_async_autoexpand(spa, spa->spa_root_vdev);
3773 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3774 	}
3775 
3776 	/*
3777 	 * See if any devices need to be probed.
3778 	 */
3779 	if (tasks & SPA_ASYNC_PROBE) {
3780 		spa_vdev_state_enter(spa);
3781 		spa_async_probe(spa, spa->spa_root_vdev);
3782 		(void) spa_vdev_state_exit(spa, NULL, 0);
3783 	}
3784 
3785 	/*
3786 	 * If any devices are done replacing, detach them.
3787 	 */
3788 	if (tasks & SPA_ASYNC_RESILVER_DONE)
3789 		spa_vdev_resilver_done(spa);
3790 
3791 	/*
3792 	 * Kick off a resilver.
3793 	 */
3794 	if (tasks & SPA_ASYNC_RESILVER)
3795 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3796 
3797 	/*
3798 	 * Let the world know that we're done.
3799 	 */
3800 	mutex_enter(&spa->spa_async_lock);
3801 	spa->spa_async_thread = NULL;
3802 	cv_broadcast(&spa->spa_async_cv);
3803 	mutex_exit(&spa->spa_async_lock);
3804 	thread_exit();
3805 }
3806 
3807 void
3808 spa_async_suspend(spa_t *spa)
3809 {
3810 	mutex_enter(&spa->spa_async_lock);
3811 	spa->spa_async_suspended++;
3812 	while (spa->spa_async_thread != NULL)
3813 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3814 	mutex_exit(&spa->spa_async_lock);
3815 }
3816 
3817 void
3818 spa_async_resume(spa_t *spa)
3819 {
3820 	mutex_enter(&spa->spa_async_lock);
3821 	ASSERT(spa->spa_async_suspended != 0);
3822 	spa->spa_async_suspended--;
3823 	mutex_exit(&spa->spa_async_lock);
3824 }
3825 
3826 static void
3827 spa_async_dispatch(spa_t *spa)
3828 {
3829 	mutex_enter(&spa->spa_async_lock);
3830 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3831 	    spa->spa_async_thread == NULL &&
3832 	    rootdir != NULL && !vn_is_readonly(rootdir))
3833 		spa->spa_async_thread = thread_create(NULL, 0,
3834 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3835 	mutex_exit(&spa->spa_async_lock);
3836 }
3837 
3838 void
3839 spa_async_request(spa_t *spa, int task)
3840 {
3841 	mutex_enter(&spa->spa_async_lock);
3842 	spa->spa_async_tasks |= task;
3843 	mutex_exit(&spa->spa_async_lock);
3844 }
3845 
3846 /*
3847  * ==========================================================================
3848  * SPA syncing routines
3849  * ==========================================================================
3850  */
3851 
3852 static void
3853 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3854 {
3855 	bplist_t *bpl = &spa->spa_sync_bplist;
3856 	dmu_tx_t *tx;
3857 	blkptr_t blk;
3858 	uint64_t itor = 0;
3859 	zio_t *zio;
3860 	int error;
3861 	uint8_t c = 1;
3862 
3863 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3864 
3865 	while (bplist_iterate(bpl, &itor, &blk) == 0) {
3866 		ASSERT(blk.blk_birth < txg);
3867 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3868 		    ZIO_FLAG_MUSTSUCCEED));
3869 	}
3870 
3871 	error = zio_wait(zio);
3872 	ASSERT3U(error, ==, 0);
3873 
3874 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3875 	bplist_vacate(bpl, tx);
3876 
3877 	/*
3878 	 * Pre-dirty the first block so we sync to convergence faster.
3879 	 * (Usually only the first block is needed.)
3880 	 */
3881 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3882 	dmu_tx_commit(tx);
3883 }
3884 
3885 static void
3886 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3887 {
3888 	char *packed = NULL;
3889 	size_t bufsize;
3890 	size_t nvsize = 0;
3891 	dmu_buf_t *db;
3892 
3893 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3894 
3895 	/*
3896 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3897 	 * information.  This avoids the dbuf_will_dirty() path and
3898 	 * saves us a pre-read to get data we don't actually care about.
3899 	 */
3900 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3901 	packed = kmem_alloc(bufsize, KM_SLEEP);
3902 
3903 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3904 	    KM_SLEEP) == 0);
3905 	bzero(packed + nvsize, bufsize - nvsize);
3906 
3907 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3908 
3909 	kmem_free(packed, bufsize);
3910 
3911 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3912 	dmu_buf_will_dirty(db, tx);
3913 	*(uint64_t *)db->db_data = nvsize;
3914 	dmu_buf_rele(db, FTAG);
3915 }
3916 
3917 static void
3918 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3919     const char *config, const char *entry)
3920 {
3921 	nvlist_t *nvroot;
3922 	nvlist_t **list;
3923 	int i;
3924 
3925 	if (!sav->sav_sync)
3926 		return;
3927 
3928 	/*
3929 	 * Update the MOS nvlist describing the list of available devices.
3930 	 * spa_validate_aux() will have already made sure this nvlist is
3931 	 * valid and the vdevs are labeled appropriately.
3932 	 */
3933 	if (sav->sav_object == 0) {
3934 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3935 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3936 		    sizeof (uint64_t), tx);
3937 		VERIFY(zap_update(spa->spa_meta_objset,
3938 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3939 		    &sav->sav_object, tx) == 0);
3940 	}
3941 
3942 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3943 	if (sav->sav_count == 0) {
3944 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3945 	} else {
3946 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3947 		for (i = 0; i < sav->sav_count; i++)
3948 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3949 			    B_FALSE, B_FALSE, B_TRUE);
3950 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3951 		    sav->sav_count) == 0);
3952 		for (i = 0; i < sav->sav_count; i++)
3953 			nvlist_free(list[i]);
3954 		kmem_free(list, sav->sav_count * sizeof (void *));
3955 	}
3956 
3957 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3958 	nvlist_free(nvroot);
3959 
3960 	sav->sav_sync = B_FALSE;
3961 }
3962 
3963 static void
3964 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3965 {
3966 	nvlist_t *config;
3967 
3968 	if (list_is_empty(&spa->spa_config_dirty_list))
3969 		return;
3970 
3971 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3972 
3973 	config = spa_config_generate(spa, spa->spa_root_vdev,
3974 	    dmu_tx_get_txg(tx), B_FALSE);
3975 
3976 	spa_config_exit(spa, SCL_STATE, FTAG);
3977 
3978 	if (spa->spa_config_syncing)
3979 		nvlist_free(spa->spa_config_syncing);
3980 	spa->spa_config_syncing = config;
3981 
3982 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3983 }
3984 
3985 /*
3986  * Set zpool properties.
3987  */
3988 static void
3989 spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3990 {
3991 	spa_t *spa = arg1;
3992 	objset_t *mos = spa->spa_meta_objset;
3993 	nvlist_t *nvp = arg2;
3994 	nvpair_t *elem;
3995 	uint64_t intval;
3996 	char *strval;
3997 	zpool_prop_t prop;
3998 	const char *propname;
3999 	zprop_type_t proptype;
4000 
4001 	mutex_enter(&spa->spa_props_lock);
4002 
4003 	elem = NULL;
4004 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
4005 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
4006 		case ZPOOL_PROP_VERSION:
4007 			/*
4008 			 * Only set version for non-zpool-creation cases
4009 			 * (set/import). spa_create() needs special care
4010 			 * for version setting.
4011 			 */
4012 			if (tx->tx_txg != TXG_INITIAL) {
4013 				VERIFY(nvpair_value_uint64(elem,
4014 				    &intval) == 0);
4015 				ASSERT(intval <= SPA_VERSION);
4016 				ASSERT(intval >= spa_version(spa));
4017 				spa->spa_uberblock.ub_version = intval;
4018 				vdev_config_dirty(spa->spa_root_vdev);
4019 			}
4020 			break;
4021 
4022 		case ZPOOL_PROP_ALTROOT:
4023 			/*
4024 			 * 'altroot' is a non-persistent property. It should
4025 			 * have been set temporarily at creation or import time.
4026 			 */
4027 			ASSERT(spa->spa_root != NULL);
4028 			break;
4029 
4030 		case ZPOOL_PROP_CACHEFILE:
4031 			/*
4032 			 * 'cachefile' is also a non-persisitent property.
4033 			 */
4034 			break;
4035 		default:
4036 			/*
4037 			 * Set pool property values in the poolprops mos object.
4038 			 */
4039 			if (spa->spa_pool_props_object == 0) {
4040 				objset_t *mos = spa->spa_meta_objset;
4041 
4042 				VERIFY((spa->spa_pool_props_object =
4043 				    zap_create(mos, DMU_OT_POOL_PROPS,
4044 				    DMU_OT_NONE, 0, tx)) > 0);
4045 
4046 				VERIFY(zap_update(mos,
4047 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4048 				    8, 1, &spa->spa_pool_props_object, tx)
4049 				    == 0);
4050 			}
4051 
4052 			/* normalize the property name */
4053 			propname = zpool_prop_to_name(prop);
4054 			proptype = zpool_prop_get_type(prop);
4055 
4056 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4057 				ASSERT(proptype == PROP_TYPE_STRING);
4058 				VERIFY(nvpair_value_string(elem, &strval) == 0);
4059 				VERIFY(zap_update(mos,
4060 				    spa->spa_pool_props_object, propname,
4061 				    1, strlen(strval) + 1, strval, tx) == 0);
4062 
4063 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4064 				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4065 
4066 				if (proptype == PROP_TYPE_INDEX) {
4067 					const char *unused;
4068 					VERIFY(zpool_prop_index_to_string(
4069 					    prop, intval, &unused) == 0);
4070 				}
4071 				VERIFY(zap_update(mos,
4072 				    spa->spa_pool_props_object, propname,
4073 				    8, 1, &intval, tx) == 0);
4074 			} else {
4075 				ASSERT(0); /* not allowed */
4076 			}
4077 
4078 			switch (prop) {
4079 			case ZPOOL_PROP_DELEGATION:
4080 				spa->spa_delegation = intval;
4081 				break;
4082 			case ZPOOL_PROP_BOOTFS:
4083 				spa->spa_bootfs = intval;
4084 				break;
4085 			case ZPOOL_PROP_FAILUREMODE:
4086 				spa->spa_failmode = intval;
4087 				break;
4088 			case ZPOOL_PROP_AUTOEXPAND:
4089 				spa->spa_autoexpand = intval;
4090 				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4091 				break;
4092 			default:
4093 				break;
4094 			}
4095 		}
4096 
4097 		/* log internal history if this is not a zpool create */
4098 		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4099 		    tx->tx_txg != TXG_INITIAL) {
4100 			spa_history_internal_log(LOG_POOL_PROPSET,
4101 			    spa, tx, cr, "%s %lld %s",
4102 			    nvpair_name(elem), intval, spa_name(spa));
4103 		}
4104 	}
4105 
4106 	mutex_exit(&spa->spa_props_lock);
4107 }
4108 
4109 /*
4110  * Sync the specified transaction group.  New blocks may be dirtied as
4111  * part of the process, so we iterate until it converges.
4112  */
4113 void
4114 spa_sync(spa_t *spa, uint64_t txg)
4115 {
4116 	dsl_pool_t *dp = spa->spa_dsl_pool;
4117 	objset_t *mos = spa->spa_meta_objset;
4118 	bplist_t *bpl = &spa->spa_sync_bplist;
4119 	vdev_t *rvd = spa->spa_root_vdev;
4120 	vdev_t *vd;
4121 	dmu_tx_t *tx;
4122 	int dirty_vdevs;
4123 	int error;
4124 
4125 	/*
4126 	 * Lock out configuration changes.
4127 	 */
4128 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4129 
4130 	spa->spa_syncing_txg = txg;
4131 	spa->spa_sync_pass = 0;
4132 
4133 	/*
4134 	 * If there are any pending vdev state changes, convert them
4135 	 * into config changes that go out with this transaction group.
4136 	 */
4137 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4138 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4139 		/*
4140 		 * We need the write lock here because, for aux vdevs,
4141 		 * calling vdev_config_dirty() modifies sav_config.
4142 		 * This is ugly and will become unnecessary when we
4143 		 * eliminate the aux vdev wart by integrating all vdevs
4144 		 * into the root vdev tree.
4145 		 */
4146 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4147 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4148 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4149 			vdev_state_clean(vd);
4150 			vdev_config_dirty(vd);
4151 		}
4152 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4153 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4154 	}
4155 	spa_config_exit(spa, SCL_STATE, FTAG);
4156 
4157 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4158 
4159 	tx = dmu_tx_create_assigned(dp, txg);
4160 
4161 	/*
4162 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4163 	 * set spa_deflate if we have no raid-z vdevs.
4164 	 */
4165 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4166 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4167 		int i;
4168 
4169 		for (i = 0; i < rvd->vdev_children; i++) {
4170 			vd = rvd->vdev_child[i];
4171 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4172 				break;
4173 		}
4174 		if (i == rvd->vdev_children) {
4175 			spa->spa_deflate = TRUE;
4176 			VERIFY(0 == zap_add(spa->spa_meta_objset,
4177 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4178 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4179 		}
4180 	}
4181 
4182 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4183 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4184 		dsl_pool_create_origin(dp, tx);
4185 
4186 		/* Keeping the origin open increases spa_minref */
4187 		spa->spa_minref += 3;
4188 	}
4189 
4190 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4191 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4192 		dsl_pool_upgrade_clones(dp, tx);
4193 	}
4194 
4195 	/*
4196 	 * If anything has changed in this txg, push the deferred frees
4197 	 * from the previous txg.  If not, leave them alone so that we
4198 	 * don't generate work on an otherwise idle system.
4199 	 */
4200 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4201 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4202 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4203 		spa_sync_deferred_frees(spa, txg);
4204 
4205 	/*
4206 	 * Iterate to convergence.
4207 	 */
4208 	do {
4209 		spa->spa_sync_pass++;
4210 
4211 		spa_sync_config_object(spa, tx);
4212 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4213 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4214 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4215 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4216 		spa_errlog_sync(spa, txg);
4217 		dsl_pool_sync(dp, txg);
4218 
4219 		dirty_vdevs = 0;
4220 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4221 			vdev_sync(vd, txg);
4222 			dirty_vdevs++;
4223 		}
4224 
4225 		bplist_sync(bpl, tx);
4226 	} while (dirty_vdevs);
4227 
4228 	bplist_close(bpl);
4229 
4230 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4231 
4232 	/*
4233 	 * Rewrite the vdev configuration (which includes the uberblock)
4234 	 * to commit the transaction group.
4235 	 *
4236 	 * If there are no dirty vdevs, we sync the uberblock to a few
4237 	 * random top-level vdevs that are known to be visible in the
4238 	 * config cache (see spa_vdev_add() for a complete description).
4239 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4240 	 */
4241 	for (;;) {
4242 		/*
4243 		 * We hold SCL_STATE to prevent vdev open/close/etc.
4244 		 * while we're attempting to write the vdev labels.
4245 		 */
4246 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4247 
4248 		if (list_is_empty(&spa->spa_config_dirty_list)) {
4249 			vdev_t *svd[SPA_DVAS_PER_BP];
4250 			int svdcount = 0;
4251 			int children = rvd->vdev_children;
4252 			int c0 = spa_get_random(children);
4253 
4254 			for (int c = 0; c < children; c++) {
4255 				vd = rvd->vdev_child[(c0 + c) % children];
4256 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4257 					continue;
4258 				svd[svdcount++] = vd;
4259 				if (svdcount == SPA_DVAS_PER_BP)
4260 					break;
4261 			}
4262 			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
4263 			if (error != 0)
4264 				error = vdev_config_sync(svd, svdcount, txg,
4265 				    B_TRUE);
4266 		} else {
4267 			error = vdev_config_sync(rvd->vdev_child,
4268 			    rvd->vdev_children, txg, B_FALSE);
4269 			if (error != 0)
4270 				error = vdev_config_sync(rvd->vdev_child,
4271 				    rvd->vdev_children, txg, B_TRUE);
4272 		}
4273 
4274 		spa_config_exit(spa, SCL_STATE, FTAG);
4275 
4276 		if (error == 0)
4277 			break;
4278 		zio_suspend(spa, NULL);
4279 		zio_resume_wait(spa);
4280 	}
4281 	dmu_tx_commit(tx);
4282 
4283 	/*
4284 	 * Clear the dirty config list.
4285 	 */
4286 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4287 		vdev_config_clean(vd);
4288 
4289 	/*
4290 	 * Now that the new config has synced transactionally,
4291 	 * let it become visible to the config cache.
4292 	 */
4293 	if (spa->spa_config_syncing != NULL) {
4294 		spa_config_set(spa, spa->spa_config_syncing);
4295 		spa->spa_config_txg = txg;
4296 		spa->spa_config_syncing = NULL;
4297 	}
4298 
4299 	spa->spa_ubsync = spa->spa_uberblock;
4300 
4301 	/*
4302 	 * Clean up the ZIL records for the synced txg.
4303 	 */
4304 	dsl_pool_zil_clean(dp);
4305 
4306 	/*
4307 	 * Update usable space statistics.
4308 	 */
4309 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4310 		vdev_sync_done(vd, txg);
4311 
4312 	/*
4313 	 * It had better be the case that we didn't dirty anything
4314 	 * since vdev_config_sync().
4315 	 */
4316 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4317 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4318 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4319 	ASSERT(bpl->bpl_queue == NULL);
4320 
4321 	spa_config_exit(spa, SCL_CONFIG, FTAG);
4322 
4323 	/*
4324 	 * If any async tasks have been requested, kick them off.
4325 	 */
4326 	spa_async_dispatch(spa);
4327 }
4328 
4329 /*
4330  * Sync all pools.  We don't want to hold the namespace lock across these
4331  * operations, so we take a reference on the spa_t and drop the lock during the
4332  * sync.
4333  */
4334 void
4335 spa_sync_allpools(void)
4336 {
4337 	spa_t *spa = NULL;
4338 	mutex_enter(&spa_namespace_lock);
4339 	while ((spa = spa_next(spa)) != NULL) {
4340 		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4341 			continue;
4342 		spa_open_ref(spa, FTAG);
4343 		mutex_exit(&spa_namespace_lock);
4344 		txg_wait_synced(spa_get_dsl(spa), 0);
4345 		mutex_enter(&spa_namespace_lock);
4346 		spa_close(spa, FTAG);
4347 	}
4348 	mutex_exit(&spa_namespace_lock);
4349 }
4350 
4351 /*
4352  * ==========================================================================
4353  * Miscellaneous routines
4354  * ==========================================================================
4355  */
4356 
4357 /*
4358  * Remove all pools in the system.
4359  */
4360 void
4361 spa_evict_all(void)
4362 {
4363 	spa_t *spa;
4364 
4365 	/*
4366 	 * Remove all cached state.  All pools should be closed now,
4367 	 * so every spa in the AVL tree should be unreferenced.
4368 	 */
4369 	mutex_enter(&spa_namespace_lock);
4370 	while ((spa = spa_next(NULL)) != NULL) {
4371 		/*
4372 		 * Stop async tasks.  The async thread may need to detach
4373 		 * a device that's been replaced, which requires grabbing
4374 		 * spa_namespace_lock, so we must drop it here.
4375 		 */
4376 		spa_open_ref(spa, FTAG);
4377 		mutex_exit(&spa_namespace_lock);
4378 		spa_async_suspend(spa);
4379 		mutex_enter(&spa_namespace_lock);
4380 		spa_close(spa, FTAG);
4381 
4382 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4383 			spa_unload(spa);
4384 			spa_deactivate(spa);
4385 		}
4386 		spa_remove(spa);
4387 	}
4388 	mutex_exit(&spa_namespace_lock);
4389 }
4390 
4391 vdev_t *
4392 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4393 {
4394 	vdev_t *vd;
4395 	int i;
4396 
4397 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4398 		return (vd);
4399 
4400 	if (aux) {
4401 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4402 			vd = spa->spa_l2cache.sav_vdevs[i];
4403 			if (vd->vdev_guid == guid)
4404 				return (vd);
4405 		}
4406 
4407 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4408 			vd = spa->spa_spares.sav_vdevs[i];
4409 			if (vd->vdev_guid == guid)
4410 				return (vd);
4411 		}
4412 	}
4413 
4414 	return (NULL);
4415 }
4416 
4417 void
4418 spa_upgrade(spa_t *spa, uint64_t version)
4419 {
4420 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4421 
4422 	/*
4423 	 * This should only be called for a non-faulted pool, and since a
4424 	 * future version would result in an unopenable pool, this shouldn't be
4425 	 * possible.
4426 	 */
4427 	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4428 	ASSERT(version >= spa->spa_uberblock.ub_version);
4429 
4430 	spa->spa_uberblock.ub_version = version;
4431 	vdev_config_dirty(spa->spa_root_vdev);
4432 
4433 	spa_config_exit(spa, SCL_ALL, FTAG);
4434 
4435 	txg_wait_synced(spa_get_dsl(spa), 0);
4436 }
4437 
4438 boolean_t
4439 spa_has_spare(spa_t *spa, uint64_t guid)
4440 {
4441 	int i;
4442 	uint64_t spareguid;
4443 	spa_aux_vdev_t *sav = &spa->spa_spares;
4444 
4445 	for (i = 0; i < sav->sav_count; i++)
4446 		if (sav->sav_vdevs[i]->vdev_guid == guid)
4447 			return (B_TRUE);
4448 
4449 	for (i = 0; i < sav->sav_npending; i++) {
4450 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4451 		    &spareguid) == 0 && spareguid == guid)
4452 			return (B_TRUE);
4453 	}
4454 
4455 	return (B_FALSE);
4456 }
4457 
4458 /*
4459  * Check if a pool has an active shared spare device.
4460  * Note: reference count of an active spare is 2, as a spare and as a replace
4461  */
4462 static boolean_t
4463 spa_has_active_shared_spare(spa_t *spa)
4464 {
4465 	int i, refcnt;
4466 	uint64_t pool;
4467 	spa_aux_vdev_t *sav = &spa->spa_spares;
4468 
4469 	for (i = 0; i < sav->sav_count; i++) {
4470 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4471 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4472 		    refcnt > 2)
4473 			return (B_TRUE);
4474 	}
4475 
4476 	return (B_FALSE);
4477 }
4478 
4479 /*
4480  * Post a sysevent corresponding to the given event.  The 'name' must be one of
4481  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4482  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4483  * in the userland libzpool, as we don't want consumers to misinterpret ztest
4484  * or zdb as real changes.
4485  */
4486 void
4487 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4488 {
4489 #ifdef _KERNEL
4490 	sysevent_t		*ev;
4491 	sysevent_attr_list_t	*attr = NULL;
4492 	sysevent_value_t	value;
4493 	sysevent_id_t		eid;
4494 
4495 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4496 	    SE_SLEEP);
4497 
4498 	value.value_type = SE_DATA_TYPE_STRING;
4499 	value.value.sv_string = spa_name(spa);
4500 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4501 		goto done;
4502 
4503 	value.value_type = SE_DATA_TYPE_UINT64;
4504 	value.value.sv_uint64 = spa_guid(spa);
4505 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4506 		goto done;
4507 
4508 	if (vd) {
4509 		value.value_type = SE_DATA_TYPE_UINT64;
4510 		value.value.sv_uint64 = vd->vdev_guid;
4511 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4512 		    SE_SLEEP) != 0)
4513 			goto done;
4514 
4515 		if (vd->vdev_path) {
4516 			value.value_type = SE_DATA_TYPE_STRING;
4517 			value.value.sv_string = vd->vdev_path;
4518 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4519 			    &value, SE_SLEEP) != 0)
4520 				goto done;
4521 		}
4522 	}
4523 
4524 	if (sysevent_attach_attributes(ev, attr) != 0)
4525 		goto done;
4526 	attr = NULL;
4527 
4528 	(void) log_sysevent(ev, SE_SLEEP, &eid);
4529 
4530 done:
4531 	if (attr)
4532 		sysevent_free_attr(attr);
4533 	sysevent_free(ev);
4534 #endif
4535 }
4536